ISSN 1580 0466 9771580046603 ECONOMIC AND BUSINESS REVIEW VOLUME 15 I NUMBER 2 I 2013 I ISSN 1580 0466 The Role of Asset Allocation Decisions in Planning for a Private Pension: The Case of Slovenia Aleš Berk Skok, Mitja Čok, Marko Košak, Jože Sambt The Persistence of Pricing Inefficiencies in the Stock Markets of the Eastern European EU Nations James Foye, Dušan Mramor, Marko Pahor Early Warning Models for Systemic Banking Crises in Montenegro Željka Asanović Shareholders' Pay-Out-Related Thresholds and Earnings Management Jernej Koren, Aljoša Valentinčič E/B/R Economic and Business Review is a refereed journal that aims to further the research and disseminate research results in the area of applied business studies. Submitted papers could be conceptual, interpretative or empirical studies. Literature reviews, conceptual frameworks and models that cover substantive insights are welcomed. Papers that employ either qualitative or quantitative research approaches, surveys, case studies, experiments or historical models can be considered for publication. It is expected that submitted articles contribute to increased understanding of the phenomenon studied and are efficiently written, methodologically clear, conceptually and empirically rigorous, readable and bias free. Authors need to highlight how the paper adds to domain knowledge or improves existing knowledge in the area of applied business research. The topics covered by the journal include (alphabetically): Accounting and Auditing, Bank and Financial Management, Business Informatics, Business Logistics, Economics, Entrepreneur ship and small business management, International Business, Management, Marketing, Money and Finance, Tourism. Today's Economic and Business Review is the successor to the Economic Review (1950-1990), the Slovenian Economic Review (1991-1999) and the Economic and Business Review for Central and South-Eastern Europe (1999-2010). EDITORS Neven Borak, Jože Damijan, Mateja Drnovšek, Jurij Jaklič, Josef Konings, Igor Lončarski,Vesna Žabkar EDITORIAL BOARD Mary Amity, Federal Reserve Bank of New York, United States Adamantios Diamantopoulos, Universität Wien, Austria Tanja Dmitrovič, University of Ljubljana, Slovenia Polona Domadenik, University of Ljubljana, Slovenia Jay Ebben, University of St.Thomas, United States Neil Garrod, University of Greenwich, United Kingdom Anja Geigenmüller, Technische Universität Bergakademie Freiberg, Germany Laszlo Halpern, Hungarian Academy of Sciences, Hungary Neven ka Hrovatin, University of Ljubljana, Slovenia Robert Kaše, University of Ljubljana, Slovenia Gay le Kerr, Queensland University of Technology, Australia Maja Makovec Brenčič, University of Ljubljana, Slovenia E/B/R Igor Masten, University of Ljubljana, Slovenia Rasto Ovin, University of Maribor, Slovenia Daniel Örtqvist, Luleà University of Technology, Sweden Marko Pahor, University of Ljubljana, Slovenia Danijel Pučko, University of Ljubljana, Slovenia John Romalis, University of Chicago, United States Friederike Schröder-Pander, VIerick Leuven Gent Management School, Belgium Christina Sichtmann, University of Vienna, Austria Sergeja Slapničar, University of Ljubljana, Slovenia Beata Smarzynska Javorcik, Oxford University, United Kingdom Jan Svejnar, University of Michigan, United States Marjan Svetličič, University of Ljubljana, Slovenia Miha škerlavaj. University of Ljubljana, Slovenia Bobekšuklev, University „St. Cyril and Methodius", Macedonia Janez Šušteršič, University of Primorska, Slovenia F it i Taki, University „St. Cyril and Methodius", Macedonia Bob Travica, University of Manitoba, Canada Peter Trkman, University of Ljubljana, Slovenia Aljoša Valentinčič, University of Ljubljana, Slovenia Irena Vida, University of Ljubljana, Slovenia Joakim Wincent, Umeà University, Sweden Jelena Zorić, University of Ljubljana, Slovenia publishers: Faculty of Economics, Kardeljeva ploščad 17, SI-1001 Ljubljana, Slovenia. The review is co-financed by Slovenian Research Agency. URL: http://www.ebrjournal.net THE REVIEW'S OFFICE: Economic and Business Review Faculty of Economics, Kardeljeva ploščad 17, SI-1001 Ljubljana, Slovenia tel : + 386 1 58 92 607, fax: + 386 1 58 92 698, email: ebr.editors@ef.uni-lj.si Bound by Birografika Bori d.o .o., Ljubljana Printed by Copi pi us d.o.o., Ljubljana Economic and Business Review is indexed in: Cabell's Directory of Open Access Journals Publishing Opportunities, DOAJ, Ebsco, Econlit, IBSS and ProQuest ISSN 1580-0466 e-ISSN 2335-4216 C / D / D Economic and Business Review is a refereed journal that aims to further the research and dissemi- ~ ' " ' nate research results in the area of applied business studies. Submitted papers could be conceptual, interpretative or empirical studies. Literature reviews, conceptual frameworks and models that cover substantive insights are welcomed. Papers that employ either qualitative or quantitative research approaches, surveys, case studies, experiments or historical models can be considered for publication. It is expected that submitted articles contribute to increased understanding of the phenomenon studied and are efficiently written, methodologically clear, conceptually and empirically rigorous, readable and bias free. Authors need to highlight how the paper adds to domain knowledge or improves existing knowledge in the area of applied business research. The topics covered by the journal include (alphabetically):Accounting and Auditing Bank and Financial Management, Business Informatics, Business Logistics, Economics, Entrepreneurship and small business management, International Business, Management, Marketing Money and Finance, Tourism. AUTHORS GUIDELINES: 1. Prepare your manuscript as an MS Word or RTF document. 2. Check to be certain the final version of the manuscript is: • Prepared in accordance with the latest edition of the APA Publication Manual. • Single-spaced throughout, including tables, figures, quotes, headings, and references. • Formatted with a running head and page number in the upper right corner of each page. 3. Abstract: Include an abstract of no more than 100 words that provides a brief and comprehensive summary of the article's contents (keep details to a minimum) as well as a good idea of the benefits to be gained from reading the article. 4. Keywords and classification: Include 5-7 keywords that accurately describe the manuscript's primary topic(s). Also include JEL (Journal of Economic Literature) Classification (http:// www.aeaweb.org/iournal/iel class svstem.php). please use no more than 2-3 descriptors. 5. Subdivision - numbered sections: Divide your article into clearly defined and numbered sections. Subsections should be numbered 1.1 (then 1.1.1, 1.1.2,...), 1.2, etc. (the abstract is not included in numbering). Use this numbering also for internal cross-referencing: do not just refer to „the text". Any subsection may be given a brief heading. Each heading should appear on its own separate line. 6. Tables and figures: Number tables consecutively in accordance with their appearance in the text. Place footnotes to tables below the table body and indicate them with superscript lowercase letters. Avoid vertical rules. Be sparing in the use of tables and ensure that the data presented in tables do not duplicate results described elsewhere in the article. 7. References: Citation in text. Please ensure that every reference cited in the text is also present in the reference list. All citations in the text should refer to: 1. Single author: the author's name (without initials, unless there is ambiguity) and the year of publication; 2. Two or three authors: all authors' names and the year of publication; 3. More authors: first author's name followed by „et al." and the year of publication. Citations may be made directly (or parenthetically). Groups of references should be listed first alphabetically, then chronologically. Examples: „as demonstrated (Allan, 1996a, 1996b, 1999; Allan & Jones, 1995). Kramer et al. (2000) have recently shown ...." Examples for reference list: References should be arranged first alphabetically and then further sorted chronologically if necessary. More than one reference from the same author(s) in the same year must be identified by the letters „a", ,,b", ,,c", etc., placed after the year of publication. Journal publication: Van der Geer, J., Hanraads, J.A.J. & Lupton, R.A. (2000). The art of writing a scientific article. Journal of the Sociological Communication, 163 (2), 51-59. Book: Strunk Jr., W. & White, E.B. ( 1979). The Elements of Style (3rd ed.). New York: Macmillan. Chapter in an edited book: Mettam, G.R. & Adams, L.B. (1999). How to prepare an electronic version of your article. In Jones, B.S. & Smith , R.Z. (Eds.), Introduction to the Electronic Age (pp. 281-304). New York: E-Publishing. Web sites Factiva (2010). Dow Jones Reuters Business Interactive LLC. http://www.factiva.com (accessed June 5, 2010). E/B/R ECONOMIC AND BUSINESS REVIEW CONTENTS 79 Aleš Berk Skok Mitja Čok Marko Košak Jože Sambt The Role of Asset Allocation Decisions in Planning for a Private Pension: The Case of Slovenia 113 James Foye Dušan Mramor Marko Pahor The Persistence of Pricing Inefficiencies in the Stock Markets of the Eastern European EU Nations 135 Željka Asanović Early Warning Models for Systemic Banking Crises in Montenegro 151 Jernej Koren Aljoša Valentinčič Shareholders' Pay-Out-Related Thresholds and Earnings Management Foreword Actions of "economic agents" are, in addition to being driven by "economic" incentives, affected by mood and emotions, as well as by other "non-economic" incentives. Prior to the "outbreak" of the last global financial and economic crisis, we experienced a relatively long period of growth and increase in prosperity, observed through increases in purchasing power and/or increased government spending. Extremely positive mood and emotions, mostly resting on two important traits of humans - salience and recency, led to a gross underestimation of "risk" and a gross overestimation of future "returns". In the aftermath of the global financial and economic crisis the tables have turned - mood and emotions have become extremely negative and sentiment is essentially determined by the perception of "risk". It is therefore not surprising that these days most of the talk, either academic, regulatory, professional, home or bar, revolves around risk (albeit in implicit terms). Financial industry does not discuss asset allocation any longer, as professionals prefer to talk about risk allocation or risk parity. Regulators are concerned with fraud risk and related financial reporting standards, with systematic risk borne by (too-big-to-fail) financial institutions and herding behavior of investors, especially bank depositors. Individuals worry about their future, about the riskiness of their savings placed either in bank deposits or entrusted to pension funds. The common theme of the papers published in this issue of Economic and Business Review is risk (albeit sometimes in implicit terms). In the first paper Berk Skok, Cok, Košak, and Sambt show the negative effects of demographics dynamics on the future pension benefits based on the pay-as-you-go system in Slovenia. Authors then continue to demonstrate benefits of private retirement savings and the important role of asset allocation decision in terms of time horizon and equity allocation. In the second paper Foye, Mramor, and Pahor investigate whether stock markets in Eastern European EU member states are weak form efficient. They show that the markets are not weak form efficient and provide possible explanations for such result. These findings have important implications for financing and investment policies of firms. In the third paper of this issue Asanovic explores the determinants of the so-called early warning models for systemic banking crisis in case of Montenegro. The author performs the analysis using Bayesian model averaging, which minimizes the subjective judgment related to the choice of early warning indicators. Finally, in the last paper of the issue Koren and Valentincic analyze to what extent do U.K. public companies engage in earnings management in order to meet dividend payout thresholds. Their findings suggest that dividend payout thresholds are significant determinants of earnings management behavior, which confirms the important role of dividends as perceived signalling mechanism. Igor Loncarski, associate editor THE ROLE OF ASSET ALLOCATION DECISIONS IN PLANNING FOR A PRIVATE PENSION: THE CASE OF SLOVENIA ALEŠ BERK SKOK1 Received: 9 January 2013 MITJA Čok2 Accepted: 2 September 2013 marko košak3 JOže SAMBT4 abstract: Current demographic dynamics driven by low fertility and increasing longevity requires adjustments of the traditional frameworks of providing pensions. In this article we highlight three crucial issues policymakers should address by implementing those adjustments. First, fiscal limitations given the current and projected demographic dynamics will dramatically reduce PAYG pensions. Without sufficient savings during the active period, individuals will increasingly end up in poverty. Their savings will not be enough to support their desired consumption in old age. Second, we highlight the impact of the asset allocation decision and the general public's related lack of awareness on this issue. Therefore, we argue that financial illiteracy about both required savings and about decisions on appropriate asset class play a significant role in determining the well-being of masses in the not-so-distant future. Third, we argue that shift towards private pension away from the PAYG is expected to come with substantial benefits stemming from diversification among conceptually different sources of pension income. Key words: PAYG, private pensions, financial literacy, old-age income, risk diversification, transition economics JEL classification: J14, G11 1. introduction Population aging requires that the traditional pay-as-you-go (PAYG) systems are downscaled. Projections of age-related expenditures from the European Commission (DG ECFIN) and Economic Policy Committee (AWG) (2009) point toward a significant risk to the sustainability of PAYG systems as a consequence of increasing demographic shifts. Muenz (2007) argues that until 2050 demographic dynamics are pro- 1 University of Ljubljana, Faculty of Economics, Slovenia, Ljubljana, e-mail: ales.berk@ef.uni-lj.si 2 University of Ljubljana, Faculty of Economics, Slovenia, Ljubljana, e-mail: mitja.cok@ef.uni-lj.si 3 University of Ljubljana, Faculty of Economics, Slovenia, Ljubljana, e-mail: marko.kosak@ef.uni-lj.si 4 University of Ljubljana, Faculty of Economics, Slovenia, Ljubljana, e-mail: joze.sambt@ef.uni-lj.si jected to result in a 10-year increase in the median age of the EU population, from 38 to 48 years old. Governments should build substantial funded pension systems as a supplement to the traditional PAYG (Du et al., 2011). More and more weight should be given to private pension systems, as under unfavourable demographic dynamics, they are far more efficient than PAYG systems (i.e., under realistic assumptions, they can deliver higher pension benefits with the same level of contributions or the same level of pension benefits with a lower level of contributions; Garrett and Rhine, 2005; Berk and Jasovic, 2007). This long-term shift toward funded private pensions should be based on sound second-or third-pillar5 frameworks, or both (Boersch Supan et al., 2008). Namely, trends in redesigning pension systems have during the past decade favoured the diversification of risks across all sources of old-age income as the coexistence of the three pillars positively effects benefits and consumption under various shocks, e.g. ageing population, inflationary shock, stock market crash etc. (World Bank Pension Conceptual Framework, 2008; Holzmann and Hinz, 2005; Lindbeck and Persson, 2003, Du et al., 2011). When a society is decreasing reliance on the PAYG and increasing reliance on private pension pillars, nature of co-movements between the drivers of pension benefits in both systems are of a great importance. Those co-movements can be measured with correlation coefficients between wages (predominant driver in the PAYG) and financial variables, i.e. stock and bond returns. Holzmann (2002) is the first published peer-reviewed research reporting very beneficial (i.e. low) national level correlation coefficients between wages and interest rates, and wages and capital return. Namely, he reports coefficients of correlation between wages and interest rates in the range between -0.197 and 0.238 and correlation coefficients between wages and capital return in the range between -0.077 and 0.202. Other authors in the area of diversification benefits report similar figures, in some cases even more beneficial (e.g. see Knell, 2010). Despite the evident shift toward private pensions, one should not expect overnight changes. Augusztinovics (2002) argues that countries, even though they redesign their pension system and move to strengthen individual pension accounts, will still deliver their pensions predominantly from PAYG systems for quite some time. Recent experience of some countries in Central and Eastern Europe provide ample evidence of the budget constraint posed by high transition costs for cases of accelerated reform towards private pensions (Simonovits, 2011). Ferber and Simpson (2009) also argue that market meltdowns make shifts towards funded pillars less politically feasible. At the same time, private pensions should not be taken for granted, as only well-managed, efficient frameworks (e.g., competitive institutions and products, broad population coverage, sound governance mechanisms) can deliver the anticipated advantages (Pensions at a Glance, 2009; Bertranou et al., 2009). 5 Second pillar includes mechanisms through which employers make contributions for their employees and third pillar the ones through which employees make their own contributions, regardless of the level of obligation. The resulting pension landscape will not only provide a more sustainable and efficient environment for managing inter-temporal consumption but also support domestic underdeveloped financial markets. Davis (2008) shows that pension-fund growth in the European Union is likely to lead to beneficial financial development with a broader range of instruments and a lower cost of capital, thus leading to higher welfare. He further argues that pension-fund growth has a significant effect on Eurozone financial markets, by moving them partly toward the Anglo-American system, as well as promoting integration. Davis and Hu (2008) provide evidence that funding improves economic performance sufficiently enough to generate resources to meet the needs of an aging population and that the improvement is even greater in emerging market economies. However, the previously mentioned characteristics of private pension systems are by themselves insufficient to provide for society's well-being if people do not have sufficient financial knowledge, i.e. are only modestly financially literate. Financial illiteracy is a very important issue, and it has been reported even for the most advanced countries (on the United States, see Lusardi and Mitchell, 2007; on the United Kingdom, see Gathergood and Disney, 2011; on Japan, see Sekita, 2011; on Germany, see Buchner-Koenen and Lusardi, 2011). Studies have found that many households are unfamiliar with even the most basic economic concepts in order to make savings and investment decisions. Financial illiteracy is lowest among women, young people, and individuals with lower incomes and lower education levels. With respect to pension savings, financial literacy increases individuals' likelihood of having a savings plan for retirement, which has a very strong impact on their wealth levels at retirement (Lusardi and Mitchell, 2007a). We argue that very important aspect of financial literacy addresses knowledge about characteristics of various asset classes for their investments. Rooij et al. (2007) found that financially illiterate individuals are significantly less likely to invest in stocks. We show in this paper that this aspect has a very significant impact on the level of pension wealth, since choosing appropriate asset classes is extremely important. Strategic asset allocation determines approximately 90 percent of portfolio performance (see Brinson et al., 1986; Ibbotson and Kaplan, 2000; Andreu et al., 2010). Overall, it is crucial that financial literacy campaigns address both topics: individuals' need to start saving for their pension (e.g., in a pension savings account) and at the same time they also need to allocate savings into appropriate asset classes. We focus on Slovenia, a country with a combination of a significantly aging population and an underdeveloped private pension system. Exclusive dataset on the distribution of individuals' income in Slovenia is used in this article to support our three main points, which are particularly important for people in countries like Slovenia who are entering a career or are halfway into their professional career. We contribute to the literature with the model, which shows the required monthly savings under each of three asset allocation choices (i.e., stocks, treasury bonds, and treasury bills). We calculate the required savings during the active work period of individuals' life under the assumption that they (together with the assumed long-term yield) can fill the gap between projected pensions from the PAYG system and the 70% net replacement rate suggest by the Organisation for Economic Co-operation and Development (OECD, 2009a). Different income levels (decile groups) are taken care of and insights into the potential outcome of a risk-aware individual allocating all of his or her pension savings into a risky diversified stock portfolio and prepares for poorly performing financial markets but actually achieving the long-term mean yield are offered. This case clearly favours investing in stocks over the long run. Finally, we address the issue of pension income diversification and show that benefits are greatest at the point, where private pension pillars only start to provide pension income. Our conclusions are relevant in general, i.e. for many developed countries across the globe, as not many current pension systems have sufficient solutions regarding an increasing old-age dependency ratio. This article is structured as follows. In the second section, we briefly describe the existing Slovenian private pension system and present pension funds in the context of the whole financial market. We also report the performance of Slovenian pension vehicles since their introduction nearly a decade ago and compare that with the performance of pension funds from developed markets. In the third section, we describe benefits from the Slovenian PAYG system and related taxation. The fourth section offers demographic projections up to 2060 and future public pension expenditures, which without changes, are expected to cause huge deficits in the pension budget. As those imbalances are unsustainable and cannot be financed through subsidies from the central government budget, we impose fiscal caps at various percentages of gross domestic product (GDP) that can be allocated to finance pensions. Those in turn pose further caps on the future levels of expected public pensions. The fifth section provides overview of three basic asset classes available for the allocation of private pension savings. Using historical data, we calculate real long-term yield and further assume that those returns are a reasonable approximation of future long-term yields. We thus use historical returns as expected returns in our model, which we present in detail in the sixth section. In section seven we present the extent diversification benefits. 2. SLOVENIAN SYSTEM OF PRIVATE PENSIONS The pension reform enacted in Slovenia in 2000 introduced private pensions within the second pillar, which comes in two forms. The first form are pensions, which are compulsory for employees in "health-risk" jobs. Employers must make special pension contributions for all such classified workers, and those contributions are transferred to employees' pension account at the special pension fund (managed by a government-sponsored institution). Second, for all other employees, participation in the defined contribution pillar is not compulsory but is promoted by a tax incentive. Namely, contributions to the second-pillar pension funds are subject to tax relief at the level of a payer. Either an employer or an employee can make a contribution, but the total amount of tax relief cannot surpass either the maximum of 5.844% of an employee's annual gross wage or a cap that is set annually6 When an employer pays a second-pillar contribution for employees, it can deduct paid contributions from the company's corporate income tax base, while in the case that a second-pillar contribution is paid by an employee, it is deducted from her personal income tax base. Table 1: Size of the second pillar, average contribution to the second pillar and breakdown of total assets at the end of 2012 MPFs PCs ICs Total AUM (mln EUR) 839.0 655.0 302.6 1,796.6 Average annual contribution 450.72 466.92 381.47 422.53 Breakdown of total assets (%)* Deposits 22.2 22.7 n.a. n.a. Government bonds 28.1 37.4 n.a. n.a. Bonds: other 29.1 32.6 n.a. n.a. Stocks 1.1 5.3 n.a. n.a. Investment funds 19.0 0.0 n.a. n.a. Cash 0.5 2.0 n.a. n.a. Total assets 100.0 100.0 n.a. n.a. Note: MPFs = mutual pension funds, PCs = pension companies, ICs = insurance companies, and AUM = assets under management; * - PC breakdown of total assets at the end of 2011. Sources: Ministry of Labour, Family and Social Affairs (http://www.mddsz.gov.si), Report on financial market trends (2013); Report on insurance market trends (2012). There were 508 thousand participants in the second pillar by the end of 2012, which represents 60.7% of the total number of persons in employment7. Different second-pillar institutions manage the individual pension accounts: insurance companies (ICs), pension companies (PCs), and mutual pension funds (MPFs). At the end of 2012 total assets under management of the second-pillar institutions was only 1,797 mln EUR, as the average annual contribution is only about 400 EUR. Assets represented only 2.1% of the assets of the overall financial sector and only 5% of the GDP (Bank of Slovenia, 2013). A notable characteristic of the Slovenian private pension system is inappropriate asset allocation. Rules about guarantees in the private pension system (Pravilnik o izračunu..., 2005) force pension managers to reach a certain percentage (at least 40%) of the cumulative yield of long-term bonds issued by the Treasury of the Republic of Slovenia on a single-member contribution. Because pension asset managers must provide additional capital in the case that their products don't deliver the guaranteed threshold yield, they do not take much risk. As a result, they tend to invest less than 5% in stocks, even though participants in the pension fund might have investment horizons extending as far ahead as 40 years. Fixed-income instruments together with 6 The cap was 2,526.2 EUR in 2008, 2,604.5 EUR in 2009, 2,646.2 EUR in 2010, 2,683.3 EUR in 2011 and 2,755.71 EUR in 2012 (Tax Administration, 2012). 7 62% we obtain using the registry data on number of persons in employment. Using the definition of International Labour Organization (ILO) the share is even lower (54%). deposits and cash represent at least 90% of total assets (see Table 1). Asset allocation in developed countries is dramatically different, as stocks represent roughly half of the total assets allocated.8 Of course, ultraconservative asset allocation can yield only meagre performance. When portfolio strategists set a conservative floor for the portfolio, the ceiling is not very high (Jensen and Sorensen, 2001). In the period 2003-2012 Slovenian pension funds recorded only 1.05% average real annual yield (mutual pension funds [MPFs]) and 0.87% (pension companies [PCs]).9 Pension vehicles as a group beat pension guarantee, as in real terms guarantee only amounted to -0.60%. Figure 1 shows the dynamics of the real yield of MPFs, PCs, the private pension system guarantee, and the best and the worst performer (in the entire 2003-2012 period) of all products in the market for the period 2003-2012 in Slovenia. If we compare same-period performance of Slovenian pension products with similar products in developed countries, we see that those countries did not have much better performance. However, there is a conceptual difference between Slovenian private pension products and those in the developed world. It is impossible to achieve long-run performance of 6-10% typical for countries with the developed private pension systems10 with the strategic asset allocations of Slovenian pension products, all of which are characterized by investment policy unification regardless of the age of their members and all of which are ultraconservative. Because Slovenia's private pension system cannot offer appropriate savings vehicles, it should change and pension products with less conservative exposure should be offered.11 Under a new system, individuals should have their choice of asset allocation—individuals have different characteristics and needs, and not all of them need a guarantee. In the "Results" section, we point out the significant impact of the asset allocation decision on the outcomes of pension savings. 8 For the end of 2009, a Towers Watson study reported the following stock allocations: Australia, 57%; Canada, 49%; Hong Kong, 62%; Japan, 36%; Netherlands, 28%; United Kingdom, 60%; and United States, 61% (2010 Global Pension Asset Study, 2010). 9 Comparison of yields between MPFs and PCs must be taken with a grain of salt, as PCs are allowed not to mark to market all of their assets. 10 Antolin (2008) reports performance between 6 and 8 percent in real terms since introduction of private pension systems, measured in geometrical terms. 11 The legislation, which allows for asset-allocation investment policy design, became effective on Jan 1 2013 (Pension and Disability Insurance Act, 2012), but the second level rules and pension products are still being prepared. F^uie 1: DyEamics of real annual yields ofMPFs and PCs in the period 2003-2012 in Slovenia (in %) MPFs PCs "Guarantee 0Best performer 0 Worst performer 25,0 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 Source: Reporton financial markettrends(2013), Repor toninsurancemarkettrends(2012), MonthlyBul-letin (2005, 2 009,22 S3(,a2Shori'cclcutation(averager with0nMPts,PC categoeies) . 3. overview of current benefits from the slovenian payg sysOeM UnOei the peieipn Saw Keing io So forsA fsom 1 JanEary ^(A^JK tEt^l aicsuol rate Nor sn SnOivEEuol roSSSi Coll letSrement LGndSSSons is EK.KL%1 Ry asKumsng Nij wiga was goow-ŠAg Sn Hal tA EtÌR the aoorRAe waRO òsi CSovenia elE replacemeAt rate amo uoAf toKN.Ee% as waII. Ths pensian bsse is cRSculfKid ai tWs iverage from tho individual's valorised best consecutive 18 years (19 in 2013, 20 in 2014 and finally 24 in 2018 and onwards). Individual's gross wages by years are transformed to nominal wages with the ratio be-twE ew aaa e a1 n e t an d g ra e e wage in that year. Those 'net' wages are multiplied with the vectorofvaloa isation noeffieients12 to calculate the pension base. Finally, accrual rate us apphnd Co the pnntion bane to cplnuPotetheamountoefirsC penclon.Ancrrial rate for menamounCsto26%0oa tdefirstlOworkingyears and furthnrC.2O%Ode Paul-additional working year. Thus, for a man with 40 working years total accrual rate is 57.25% (26% + 1.25% * 25 years). For women the pension system is more generous with 29% for the first 15 years and, again, 1.25% for each additional working year. Thus, for women with 12 Calculation of those factors was based on the past growth of pensions relative to wages. After the pension law introduced with 1 January 2013 the set of factors will grow in line with the growth of average wage - thus, it will not depend on the growth of pensions any more. In the past factors were declining because the pension growth was lagging behind the wage growth. 40 working years the total accrual rate of 60.25% is applied. From 2013 to 2022 there is a transition period in which less than 40 working years is required for women and therefore in this period for women even higher accrual rate is applied for 40 working years13. However, in our calculations we focus only on male with 40 working years. Slovenia, as other countries from Central and Eastern Europe, has undergone through several phases of pension reform; the last phase had passed the Parliament in 2012 and is effective from January 2013. Its most important element is a gradual increase of retirement age for both genders. The full retirement age (for old age pension) is thus increasing from 61 years (women) and 63 years (men) to 65 years (by 2016 for men and by 2020 for women). However, under both pension systems earlier retirement (up to several years) was/is possible with full benefits and without penalties if the person collects required number of working years earlier. Slovenia is also characterised by the fact that it has never implemented a compulsory second pillar14 compared for example with Slovakia, Hungary and Poland. However, the compulsory second pillar has been recently effectively abandoned in Hungary, while in Slovakia it is not compulsory any more since February 2013. The Czech Republic which initially also did not introduce mandatory second pillar is now opening the option for employees to divert part of their contributions from the first to the second pillar (Berk et al., 2013). 3.1. Financing the PAYG pillar Compulsory pension contributions for the PAYG pillar are set at the rate of 24.35% (employees pay 15.5%; employers, 8.85%) out of a gross wage without any ceiling.15 The aggregate contributions total 3,348.9 million EUR, or 9.5% of GDP, in 2012. Because this is not sufficient to cover expenditures of the first pillar (which totalled 4,851.0 million EUR, or 13.7% of GDP, in 2012) in aiming to maintain the financial stability of the system, current legislation has stipulated that the central government budget cover the rest. In 2012 that transfer amounted to 1,416.2 million EUR, or 29.2% of total PAYG revenues (Ministry of Finance, 2013). 3.2. Taxation of Pensions Contributions for the PAYG pillar are entirely deductable from the personal income tax base, while pensions from the PAYG pillar are subject to personal income tax under 13 For 40 years of work women receive total accrual rate of 64.25% if they retire in 2013-2016 period, 63.5% in 2017-2019 and 61.5% for retiring in 2020-2022 period. Nevertheless, minimum and maximum pension base as of December 2012 are 551.2 EUR and 2,204.4 EUR, respectively. Taking into account that there is no ceiling for the PAYG contributions, such a pension base setting mechanism has a strong redistributive effect. 14 Exemptions are some selected professions, such as miner, or soldiers, where additional compulsory contributions paid by employers' are collected by special government owned pension fund. 15 The self-employed on the other hand pays the same rate of contributions from the base which is a function of annual income from self-employment with the ceiling equal to 2.4 average national gross wage. an advantageous tax-credit system. As a result, most pensions (approximately 97%) are effectively tax-free, whereas the remaining 3% are taxed at a relatively low effective tax rate. On the other hand, the contributions to the second pillar are deductable from the personal income tax base up to the certain level. This tax relief is limited with the 5.844% of employee's annual gross wage or the nominal amount set annually (2,755.7 EUR in 2013) - whatever it is lower. Pensions from the second pillar are not entitled to the same tax credit as pensions from the first pillar. Instead, 50% of the second-pillar pension is subject to tax, without any special tax credit. As a result, these pensions are taxed more than the first-pillar pensions. Table 2 includes average gross and net wages in 2013, as well as the maximum amount of tax relief for the second-pillar contribution. One can observe that only taxpayers from the highest decile group can take full advantage of the nominal tax relief for second-pillar contributions. Table 2: Gross average annual wage, net average annual wage and maximum amount of tax relief for the second-pillar contribution in 2013 (in EUR) Decile group Average gross wage Average net wage Maximum tax relief (5,844 %) 1 9,671.5 6,455.4 565.2 2 10,760.0 7,140.1 628.8 3 11,939.4 7,875.8 697.7 4 13,190.8 8,642.1 770.9 5 14,563.9 9,445.3 851.1 6 16,288.2 10,417.1 951.9 7 18,546.8 11,661.9 1,083.9 8 21,740.3 13,385.6 1,270.5 9 26,755.5 15,872.7 1,563.6 10 47,663.8 24,743.2 2,785.5 Source: Authors' calculation based on data from Statistical Office (2013). 4. THE IMPACT OF DEMOGRAPHIC CHANGES ON BENEFITS FROM THE PAYG PILLAR The twentieth century experienced explosive population growth, but the twenty-first century is likely to see the end of population growth and instead population aging (Lutz et al., 2004). According to population projections, in the future there will be strong demographic pressures on public expenditures for pensions, health care, and long-term care (European Commission, 2012). Scholars began warning of this decades ago, but we have seen no changes, mainly because short-term-oriented politicians have as their horizon only the next elections. They are not interested in projections for a distant future. The situation, though, has become so aggravated that taking action cannot be further postponed. Many countries have already taken various measures. International organizations are pressuring countries to act in a timely manner to facilitate and accelerate change. PAYG systems are vulnerable to population aging. In our analysis, we apply Eurostat population projections from EUROPOP2010 for 2010-2060. They were prepared by the Eurostat for the European countries (EU27) and European Free Trade Association countries (EFTA)16. The projections assumed gradual convergence of countries' mortality and fertility, with the year 2150 set as the convergence year. However, the projections were prepared only until 2060, when only partial convergence has been reached. In Slovenia the life expectancy at birth is increasing rapidly. The past decade alone (from 2000-2001 to 2011) saw an increase of almost 4.5 years for males (72.1 to 76.6 years) and 3.3 years for females (from 79.6 to 82.9 years) (Statistical Office of the Republik of Slovenia, 2012, p. 79). Some developed countries already have a considerably higher, and still-increasing, life expectancy than Slovenia.17 The current population age structure is given. The baby-boom generations, born after World War II during times of high fertility, are now in their 50s and early 60s. Over the coming decade, they will be intensively entering retirement. At the same time, people born during the 1980s are starting to enter the labour market. During the 1980s and 1990s, fertility declined; in the first half of the 2000s, it stabilized at very low levels. In 1980 total fertility rate (TFR, or the average number of children a woman gives birth to, during her fertility period) was 2.1, which was still a replacement-level fertility. Since then, TFR declined until 2003, when it reached only 1.2 (Statistical Office of the Republic of Slovenia, 2008, p. 56). Consequently, the number of newborns decreased sharply in that period. In 2003 just 17,321 children were born in Slovenia, whereas the figure was 30,604 in 1979 (Statistical Office of the Republik of Slovenia, 2012, p. 78). Those reduced generations (they are only about one half of their parent's generations) will also determine fertility levels in the coming two to three decades. Even if fertility (TFR) were to increase, which the projections assume, the absolute number of newborns is expected to fall considerably because there will be fewer women of reproductive age. Sensitivity analysis (in which we variate fertility assumptions while keeping other assumptions unchanged) shows that, despite the impact of fertility on population size in the long run, we cannot expect increased fertility to considerably mitigate the process of population aging in Slovenia in the coming decades (Sambt, 2008). Further, from an economic point of view, increased fertility does not have positive economic 16 Iceland, Liechtenstein, Norway and Switzerland. 17 E.g., in Japan the life expectancy at birth in 2009 was 79.6 years for males and 86.4 years for females (OECD, 2011). effects for about 20 years, as cohorts of newborns start to enter the labour market. In the meantime, the economic effect can even be negative, causing higher public expenditures in the form of education and other transfers like child allowances and health care. Immigration decreases the overall aging of the population, especially because most immigrants are relatively young (Eurostat, 2011). However, without assuming unreasonable high immigration, the positive effect is only moderate. With time, immigrants are also aging and entering the age group of 65 and older (Bonin et al., 2000). Figure 2 presents the projected dynamics of the age structure of the Slovenian population by three broad age groups related to economic activity:18 0-19, 20-64, and 65 and older. According to EUROPOP2010 projections by 2060 the Slovenian population should slightly increase - by 11,000 people, which is 0.5% of the total population. However, the change in the age structure of the population is striking. The percentage of people age 65 years and older is expected to almost double in the 2012-2060 period, from 16.6% to 31.6%. In contrast, the size of the working-age population (age 20-64) is expected to shrink considerably - from 64.3% in 2012 to 49.8% in 2060. The combination of those two processes will have serious consequences for the long-term sustainability of public finance systems, unless adjusted accordingly. Sensitivity analysis reveals that those results are very robust for a broad range of assumptions about fertility, mortality and migrations since they are mainly driven by the increasing longevity, and especially by the given population structure (Sambt, 2008, Sambt, 2009). The unfavourable economic development with respect to the population age structure can be shown with the old-age dependency ratio, which is calculated as the ratio between the elderly (age 65+) and the working-age population (age 20-64), multiplied by 100. An increasing old-age dependency ratio indicates an increasing demographic burden on the productive part of the population in order to maintain the pensions of the economically dependent. According to the EUROPOP2010 population projections, the old-age dependency ratio in Slovenia will increase from 25.9 in 2012 to 63.4 in 2060 (see Table 3). A rapidly increasing old-age dependency ratio is not specific only to Slovenia. Practically all developed countries across the globe face strong population aging. Therefore, the analysis we present here is generalizable. Table 5 presents projected future increases in the old-age dependency ratio for all EU27 member states, including Slovenia. In the new EU member states (EU12), the increase is expected to be somewhat stronger than in the old EU member states (EU15). 18 In demography traditionally defined dependency ratio compares population aged 65+ with population aged 15-64. However, in developed countries using 20-64 years in the denominator has been seen as more adequate from the economic point of view since not many individuals enter the labour market before age 20. Figure 2: Slovenian population in broad age groups: EUROPOP2010 projections for 2012-2060 (%) Source: Eurostat, 2011. Tabk3: ORVvagedepeNdsncyraSioinEUcountsLesiEOROPOPOOlOproScstLons^for2002, 20CEVND2060 Old memOyr states(E U15) newmomU9is tale s dVi N 2030 2060 2012 2030 2060 Belgium 29.2 40.k ken Bilgaria ee.7 40.4 26.3 Denmark 29.6 46.7 48.0 CzecliRepublic 2ee 37.6 rco.e Germany 33.a 21.0 65.1 Estonir 07.B 2 94 61.4 Spain 27.6 38.6 61.6 Greece 31.9 41.9 62.2 France 29.4 43.4 51.7 Latvia 27.4 39.4 74.3 Italy 34.0 44.5 64.5 Lithuania 26.0 38.8 62.3 Cyprus 21.7 33.7 48.7 Hungary 26.8 36.4 62.8 Ireland 19.9 31.2 48.5 Malta 25.8 42.8 60.5 Luxembourg 22.5 32.7 43.0 Poland 21.3 38.6 70.5 Netherlands 26.8 44.2 51.7 Romania 23.3 32.8 70.4 Austria 28.7 42.1 55.4 Slovenia 25.9 42.5 63.4 Portugal 29.9 40.9 59.7 Slovakia 19.0 34.2 67.4 Finland 30.6 47.3 54.5 Sweden 32.3 41.3 51.7 UK 28.6 38.6 46.5 Source: Eurostat, 2011. 4.1. Projecting Future Public Pension Expenditures Strong population aging translates into pressure on the public pension system. The model that we use in the simulations rests on the age profiles from the base year. Therefore, we refer to it as the age-profiles-based model. Such models are standard approach in Generational accounting methodology19. This model is also used for projecting pension expenditures for Slovenia published in 'The Ageing reports' by the European Commission (Ageing Working Group). In calculations we use three types of matrices. The matrix of pension age profiles (PENS) includes average pensions by years in the future. It builds on the pension age profile from the base year (2011). In particular, the PENS matrix consists of two matrices multiplied with each other. The first one contains age profiles of average pension benefits per receiver, whereas the second one includes the share of pensioners in the total population by age group (i.e., retirement rates). This decomposition enables us to more easily, and more accurately, introduce future changes into the age profiles (e.g., increasing retirement age). Every year those age profiles are shifted up by the assumed growth of pensions. Before 2013 all pensioners with the same retirement conditions received the same level of pension, regardless of the time of retirement ('horizontal equalization'), which strongly simplified the calculations. Horizontal equalization in Slovenian pension system was achieved through complex mechanism of valorisation that was abolished in 2013. Now growth of pension is in real terms20 60% indexed to the growth of wages. We follow each cohort of pensioners separately. We use the standard macroeconomic assumption that wages grow in line with labour productivity growth - we use the latest European Commission assumptions. The coefficient matrix (C) summarizes the effects of future departures from the basic age profile, assumed in the pension matrix. It contains the impact of the Slovenian pension system on pension age profiles in the future. The legally enforced, but gradually phas-ing-in parameters of the Slovenian pension reform are a typical such case. We obtained several inputs for the coefficient matrix (C) from simulations on microdata on pensioners who have already retired. We simulate their behaviour under pension parameters that will be valid in future years. Weighted averages of those results (by age groups) enter the coefficient matrix. The population matrix (P) contains the EUROPOP2010 population projections presented earlier. 19 For review of Generational accounting methodology see, for example, the initial paper from Aurebach, Gokhale and Kotlikoff (1991) and comparative studies across countries (European Commission, 2000; Auerbach, Kotlikoff, & Leibfritz, 1999). 20 Formally, the growth of pensions is 60% indexed to the nominal growth of wages and 40% to the growth of consumer price index (inflation). This is equivalent to 60% indexation in real terms. The amount: of p vnvion e:xpenditu.ren on ind mduek aged! k in ye vr y is thu s c okutó ed as follows (rnatriNes are multip>lieit in an element-by-eldmeni manneop PENSEXPat, = PENS^C^G^ (1) where G cN^Vo^nvc it^ff^ifS^ntvof ohe Efneivnilsnm tfin^^^ee^i^ae (here, 2ENNS io time t. Ppthc pvnion v:p]3snel^turee; |f^kSfSEi^;ui iu: s^var f afe cakulateid as she pum of gNO-eetetl i^xaendiSuree tlivaeneiatt by aHage groups: D PENSEXPt = Y, PENSEXPat (2) a=0 where index a auns fnem 0 to D,and D Vnnel ecthei^ax^^umiee- Eh of l(fv on o uo model, tibi eye geoup fOO unii older). Finally, we employed the setoo ue aeNoeeNnvmin uasompRions from ti^^ European Com-mOvsion (sef 2ind in VNilp inglupiag its^umf tine v on joeoduttf ivil^^ g-!euilh, BB-Dfp emolef-mintl nbUvnemp1oymfnlrales2O In -he future the agrprofiieoO Emglovmyntra0erisnr°jevlud^o^hi^ inloh ighor ages. Cnnspqueetl., alio ngf profile pa retirement sotoo wlli wilhdpaw inluliigOre ggok. Broh of tdofn two eifuola aie enleriuu ilie made) IhrouuE nWNS mv0rix.)nrreksinu endo plooJ-nEnl aaton (alio tecanse of ilrpumaV duciine iu vnampleymenl rotevi wiU home pr ni-ttoe insn^^t also uro GDP blnne GDl onaw-e in Sim modal aouwiits ualp0lour proUuclrai Gy el^/") O fN VO >n o tN O © tN Year Source: authors' calculations. There are basically three solutions for mitigating rapidly growing pension expenditures. The first, usually considered preferable, is to increase the retirement age. This solution also provides the most straightforward response to increasing longevity. The second solution is to increase taxes, usually on labour income. In Slovenia labour is already highly taxed, which hinders its international competitiveness. Moreover, the tax burden has a negative impact on employment and incentives to work. The third solution is to reduce the level of pension benefits from the PAYG system. In our analysis we focus on the third option by introducing a simple assumption about future reductions of pension benefits. We assume the government will have to prevent further increases in public pension expenditure above some percentage of the GDP (i.e., capping expenditures) in a way that all pensions will be cut proportionally, regardless of the type and level of pension. Thus, we set the tolerated maximum percentage of public pensions of GDP at 11%, 12%, 13%, 14%, and 15% of GDP. Figure 3 shows projected public pension expenditures as a percentage of GDP according to these scenarios. 4.2. Expected Level of Pensions from the PAYG Pillar As already explained in section 3 in 2013 the statutory accrual rate for a man with 40 working years was set to 57.25% and according to the current pension law it will remain at this level in the future ('No limitations' scenario in Figure 4). We also present results for different scenarios of assumed maximums to which government would tolerate pension expenditures to increase; in those cases, the net replacement rate would fall to even (much) lower levels, as it is revealed in Figure 4. In the case that expenditures for pensions would be capped at 11% of GDP, the net replacement would thus decline to 37% by 2060. Figure 4: Projections of net replacement rate at retirement in the period 2012-2060 ro CN r— LO CN LO LO CO CO CO CO CO CO o LOLOCOOt— LO'— CN

% -S S ^ e 'S S rt ^ T3 s ECONOMIC ANDBUSINESS REVIEW| 3 due to high kurtoois, rathto than skewness, thtsefosu ths jearamu.riE models we apply still eoturn oobusl .esulto. U METHODOnUGY The Seste wo employ fpll inio lowocapegories: iesir ob sui'isO independonces uniU rooetests, multifile parianpu iutio Ursts snU liquiisSity. "We chose to eephoaie the meihodelogy of i/VorSliéngton anci Hig°s (2h0nS for ehr rrsiat mdogendeiK^ ooü: sc^nt teste sisiel muetì]ule gar ianre ratui^ Oestoe bocaust oS1 the tireaid euige ofWFMU°eotEappliedEytheeethors and 0 he e euE |seiltto o et leo^egeel in thelitora0 ure. Geeffin et al (° OW) psetrungesi:'°ests similar to tOeones emjiloyed iin toglie pajtee, Shet quoutioa rfsSEthrr it io feasüulp tomake state-mealo eSaud rsla^iii.0«; matket rffidtneu iniernationalio uukrr orse cat euihaol (or Ihr in-eormaelon tetr^irontne^tl WUiis oui daieget sovdrs nstions wrtO eietlsidlianSčc i similarities, uee sunn or ruCe oui: hies poasif éltiy 0har oet cts ee hs Covi decn tirtoelor e°y Oins.Whüe opr tlaiaest coilras et kege gcogegpUtc aria, Che majoeity oeeiscks fae ssanatpct osa Stie ^o^sitov ptoEk exthangS1 To aontrrteoB sna Vohshbisis, wc prefoim lituis Oesta foe boOh 1/itL e segéun rc a wie tr i e oeiP the ineh mdp al ere ntii a oSa a. 1 Tesio oU airiat itedependgnou W ttasa o ee rirt as e aid to bt e ^eiaCCrie c onre i aied oU f re^et sta o oli o timr serie e of ser urns wii ir 10s osen loes yioiess2otietisa^yBianificane teopite: E(Ayilt\Ay^) = ß± + ßzAyu-! WWerei E^y^by^) = theexi°ecled vaUus of Ayit given Ayit_1 ßl = theregression intercept b2 = the regression slope Unlike serial correlation, the runs test is non-parametric and therefore does not require thereturnstobe normallydistributed.Runs tests determinewhetheratime seriesfol-towse random walkbycountingthenumber ofconeecubive pasétiveotneaariveopterva-CibdsanCcombariugiUtoan eopected valpe (E(R)): , N N + 2NuNd E(R) =-— V y N Where: N = Numberofobservations N = Numberofpositiveo bservations Nd = Numberofnegativeobservations R = fumbfr ofouns We use the expected value a nd variance values (V(R)) to calculate a teststatlstic.Z: V(^=2NUNP (2NP N ~N) ( ) (N)2(N-l) R - E(R) Z= ' VV(R) The null hypothesis is that the returns can be considered to follow a random walk process. Rejection of the null hypothesis indicates that the stock's returns are non-random and contravene WFME. In order to test whether EU accession resulted in an increase in WFMEi we ase a z-test to Satetmine ifthe percentage ofsta pks considered statistically stgnificanl at a pacticdZar digniScance fevnl iz^^^tSsticzlly tèffenent between the pre- and O ectias cessirn datcdcc- . 3.2 Unit root tests Unit root tests are used to determine whether the log returns of stocks in our dataset is stationary, i.e. whether it has constant statistical properties; if stocks follow a random wc skprocess, a tock cE^ic c Et choc Id b e non-ntutionar y. We uzeAr a e variants, Augmented DickeyFullrf-ADF^Phülipc - Paca on(fP) a nd KwaitokowH.Phillips, S chmidt and Shin -KCbCi. ADS i-lbn mosS wclf-kdcwb udtt roditssl, thenull hypalhcsis is that the data is nonsta- t^c^iba^^.ltz^rb^^^i^^e^c calnulotcdCy risonlngiheCniiuwingregression: q Ayit = ßo + ß1tr + a^t-i + ap ^ ^Vit-p + £pt p=i Where: a = the coefficients to be estimated q = number of lagged terms b0 = intercept b = trend coefficient tr = trend MacKinnon's critical values are then applied to determine the significance of a. The PP test, developed by Phillips and Perron (1988), extends ADF to allow errors to be independent and heteroscedastic. For a complete derivation, see Phillips and Perron (1988). WhiletheADF and PP testsh lve r^i^Hti^pc^t h^^^^js of nonstationarity, the KPSS test has a nullhypothesisofstationarity. Repeeeingthe nulldrrothesis provides a useful validation check for the results from the ADF and PP tests. The reader should consult Kwiatkowski etal.(lP92)for etullPefivatioy.As withthefesfsof re rial independence, we apply a z-teetlodeptrmins erPotest ^lufsi^fu l^^fi^e^enth^ysu- foci post-accession datasets can be cynritleesi thatirti ca^U^erene 3.3 Multiple Variance Ratio Tests The third set of statistics employed are multiple variance ratio (MVR) tests. This approach wardevelopenUn Lo md MadVšnlpy((9PneonUand Chow and Denning (1993) who coArtructed theMVR Ueete Snorder todetect I(^) h autocorrelation and heteroscedastic- itrr^jrr^^ye nf.^UpicSrimpo ftan(decpuuaifttockr doi low a random walk, the variance of returnr rhould rire ar a linear function to the number of obrervationr. That ir, the vari- z z ance ratio of the returnr over qq period murt be equal to qu q& . The variance ratio (VR) ir calculated ar: VR(q) = 2 T 2 o 2(q) o 2(i) Where: G2(1) = variance of daily log returns q = number of periods used for the sampling interval 02(q) = (1/ q)multiplied by the variance of q-daily returns If stocks conform to the random walk process, VR should not be statistically different to one. In line with the methodology of Worthington and Higgs (2004), the sampling intervals used for q were 2, 5, 10 and 20 days. For a more in depth overview of MVR methodology or a complete derivation, the reader should consult Worthington and Higgs (2004) or Chow and Denning (1993) respectively. We also apply a z-test to determine whether the pre-andpost-accession results are statisticallydifferent. 3.4LiquidityControls Studies frequently conclude that liquidity is related to future returns. Examples of such work include Amihud and Mendelson (1986, 1989), Chordia et al (2001), Jones (2002), Amihud (2002), and Brennan et al (1998). Datar et al (1998) demonstrate a negative correlation between liquidity, as measured by turnover, and returns. Haugen and Baker (1996) found that liquidity is one of several generic factors that explain returns across global stock markets. Brzeszczynski et al (2011) found that trading intensity affected beta calculations for stocks listed on the Warsaw Stock Exchange and thus had serious ramifications for corporate finance decisions. The relatively small size of the stock markets of the EE EU countries raises the concern that our results could be distorted by liquidity issues. Liquidity is an elusive concept, consequently in Table 5 we employ three widely used measures to control for it: i) Market capitalization ii) Average volume divided by shares outstanding iii) Bid-ask spread divided by share price. We create liquidity portfolios by assigning a rank (1 (low) to 5 (high)) to every stock for each of the three liquidity measures. Then we separate the combined results from Tables 2, 3, and 4 into five liquidity ranked portfolios in order to examine the effects of liquidity on the tests employed; we repeat this for each of market capitalization (Panel A) average volume divided by shares outstanding (Panel B) and Bid-ask spread (Panel C). 4. RESULTS The results from the tests of serial independence, unit root tests and multiple variance ratio tests are shown in Tables 2, 3 and 4 respectively. As we cover a large geographic region, each table also provides a geographic breakdown of the results. While around one-third of our dataset is listed outside Poland, the shares are listed on a lot of different exchanges; no exchange other that the Warsaw Stock Exchange has more than 14 shares in the dataset. This makes inferences for individual countries difficult. 4.1 Tests of serial independence Table 2 shows the results from the tests of serial independence, the serial correlation coefficient and the runs test. Looking at all the stock exchanges in the dataset, even at the 0.01 level of significance, almost one third of the stocks in our dataset return significant t-statistics from the serial correlation regressions for both the pre- and post-EU accession periods. Whilst there has been a marginal decrease in the number of stocks statistically significant at the 0.01 level between the pre- and post-accession datasets, the z-test reveals that the difference is not statistically significant. 43% of stocks in our dataset can be considered serially correlated at the 0.1 significance level for the pre-accession period; this rises to 66% for the post-accession period. The z-test reveals that the increase in the number of stocks exhibiting serial correlation at the 0.05 and 0.1 levels is statistically significant at 0.01, indicating that prices of stocks listed in the EE EU nations may have actually become less efficient. Looking at the individual stock exchanges, it can be seen that the results from the stock exchanges of other countries are largely consistent with those from the Warsaw Stock Exchange. Across the majority of stock exchanges most stocks exhibit properties consistent with serial correlation, at least at the 0.1 level. The z-test reveals no statistically significant difference between the pre- and post-accession datasets. Thus we can comfortably reject the null hypothesis that returns in the stock markets of the EE EU are not serially correlated. Table 2: Tests of Serial Independence Serial Correlation T Statistic 1999-2003 2004-2008 Z Test 1999-2003 Runs Test 2004-2008 Z Test Entire Region % of Observations Significant at % of Negative Observations Czech Republic % of Observations Significant at % of Negative Observations Estonia % of Negative Observations Hungary % of Observations Significant at % of Negative Observations Latvia % of Observations Significant at % of Negative Observations Lithuania % of Observations Significant at % of Negative Observations Poland % of Observations Significant at % of Negative Observations Slovakia % of Observations Significant at % of Negative Observations Slovenia % of Observations Significant at % of Negative Observations 1% 31% 28% 0,41 22% 19% 5% 39% 54% - 2,23 38% 38% 10% 43% 66% - 3,45 46% 49% 15% 42% 64% 69% 1% 40% 29% 20% 0% 5% 60% 43% 40% 29% 10% 60% 71% 60% 43% 0% 29% 80% 57% 1% 60% 50% 40% 38% 5% 60% 63% 60% 50% 10% 60% 63% 60% 63% 60% 75% 20% 63% 1% 13% 38% 13% 38% 5% 38% 50% 13% 63% 10% 50% 63% 13% 75% 38% 50% 13% 38% 1% 0% 100% 0% 0% 5% 0% 100% 0% 0% 10% 0% 100% 0% 0% 0% 0% 100% 0% 1% 40% 70% 70% 30% 5% 70% 90% 90% 50% 10% 70% 90% 90% 70% 30% 20% 90% 90% 1% 13% 16% 11% 18% 5% 16% 48% 27% 35% 10% 22% 63% 40% 44% 11% 48% 71% 74% 1% 100% 0% 100% 100% 5% 100% 0% 100% 100% 10% 100% 0% 100% 100% 0% 0% 0% 100% 1% 100% 71% 25% 7% 5% 100% 71% 50% 36% 10% 100% 71% 50% 50% 0% 0% 58% 50% 0,47 - 0,04 - 0,40 All calculations are based on stock returns calculated on natural logarithms of Bloomberg last prices in local currencies. Serial correlation is calculated using one day lags Runs tests calculations are based on the sign of returns When the runs test was applied to our dataset, about one fifth of stocks yielded statistically significant results even at the most stringent 0.01 level for both the 1999-2004 and 2004-2008 datasets. Around half of both the pre- and post-accession datasets can be considered significant at the 0.1 level. Stocks listed on the Riga Stock Exchange perform poorly in the runs tests, but the dataset only contains one stock from this country; excluding Latvia, the non-Polish stock markets have similar results to the entire dataset. 4.2 Unit root tests Table 3 shows the results from the three sets of statistics that form the unit root tests. The null hypothesis of the ADF and PP tests is that the time series has a unit root. The KPSS test reverses the null hypothesis and assumes that the time series has no unit root. Both the ADF and PP tests reject the null hypothesis, even at 0.01, for all stocks in both the pre- and post-accession datasets. We can comfortably reject the null hypothesis of nonstationarity for all stocks. Needless to say, there is no country variation here. Both tests clearly indicate that the returns of all stocks in the dataset are stationary, that is follow a deterministic rather than stochastic trend; inconsistent with a random walk. Out of all the metrics we employ, only the KPSS test indicates that stationarity may have declined between the pre- and post-accession periods. The KPSS statistic is insignificant for less than half of all stocks at the 0.01 level of significance for the post-accession dataset, indicating that we cannot reject the null hypothesis of no unit root; yet for our pre-accession dataset, only 5% of stocks have KPSS statistics that can be considered statistically significant at the 0.01 level. Whilst almost three quarters of post-accession stocks have KPSS statistics that can be considered statistically significant at 0.1, the corresponding figure for the pre-accession nations is only around one quarter. The z-test reveals that there is a statistically significant increase in the KPSS statistic between the pre- and post-accession datasets. The results from Poland are almost identical to those for the region as a whole, indicating little regional variation. While the KPSS statistic is less conclusive than ADF or PP, we can still confidently infer that all three unit root tests employed indicate that returns of many stocks listed in the EE EU nations are stationary, leading us to reject the null hypothesis that stocks follow a random walk. 4.3 Multiple Variance Ratio Tests Table 4 shows the results from the MVR tests using sampling intervals of two days, 5 five days, 10 days and 20 days; corresponding to one day, one week, one fortnight and one month. Table 3: Unit Root Tests ADF_Phillips-Perron Test_KPSS Test 1999-2003 2004-2008 1999-2003 2004-2008 1999-2003 2004-2008 Z Test Entire Region % of Observations 1% 100% 100% 100% 100% 5% 46% - 6,81 Significant at 5% 100% 100% 100% 100% 13% 64% - 7,86 10% 100% 100% 100% 100% 25% 72% - 7,31 Average -29,27 -28,88 -33,76 -31,09 0,26 0,79 Absolute Average 29,27 28,88 33,76 31,09 0,26 0,79 Czech Republic % of Observations 1% 100% 100% 100% 100% 0% 0% Significant at 5% 100% 100% 100% 100% 0% 14% 10% 100% 100% 100% 100% 20% 29% Average -31,61 -28,49 -33,00 -32,15 0,16 0,34 Absolute Average 31,61 28,49 33,00 32,15 0,16 0,34 Estonia % of Observations 1% 100% 100% 100% 100% 0% 63% Significant at 5% 100% 100% 100% 100% 0% 75% 10% 100% 100% 100% 100% 20% 75% Average -31,57 -28,50 -34,43 -30,91 0,24 0,88 Absolute Average 31,57 28,50 34,43 30,91 0,24 0,88 Hungary % ofObservations 1% 100% 100% 100% 100% 0% 50% Significant at 5% 100% 100% 100% 100% 0% 75% 10% 100% 100% 100% 100% 0% 75% Average -31,32 -31,86 -31,37 -35,00 0,13 0,65 Absolute Average 31,32 31,86 31,37 35,00 0,13 0,65 Latvia % of Observations 1% 100% 100% 100% 100% 0% 100% Significant at 5% 100% 100% 100% 100% 0% 100% 10% 100% 100% 100% 100% 0% 100% Average -36,60 -35,06 -36,58 -35,10 0,27 1,04 Absolute Average 36,60 35,06 36,58 35,10 0,27 1,04 Lithuania % of Observations 1% 100% 100% 100% 100% 30% 90% Significant at 5% 100% 100% 100% 100% 60% 100% 10% 100% 100% 100% 100% 60% 100% Average -18,52 -25,24 -30,20 -32,16 0,56 1,73 Absolute Average 18,52 25,24 30,20 32,16 0,56 1,73 Poland % of Observations 1% 100% 100% 100% 100% 4% 39% Significant at 5% 100% 100% 100% 100% 9% 60% 10% 100% 100% 100% 100% 22% 69% Average -30,73 -28,75 -33,88 -30,30 0,23 0,71 Absolute Average 30,73 28,75 33,88 30,30 0,23 0,71 Slovakia % of Observations 1% 100% 100% 100% 100% 0% 100% Significant at 5% 100% 100% 100% 100% 0% 100% 10% 100% 100% 100% 100% 0% 100% Average -22,35 -34,85 -22,24 -34,81 0,11 0,96 Absolute Average 22,35 34,85 22,24 34,81 0,11 0,96 Slovenia % of Observations 1% 100% 100% 100% 100% 0% 64% Significant at 5% 100% 100% 100% 100% 17% 79% 10% 100% 100% 100% 100% 33% 93% Average -28,19 -30,21 -38,55 -32,87 0,27 0,93 Absolute Average 28,19 30,21 38,55 32,87 0,27 0,93 All calculations were made on natural logarithms of Bloomberg last prices in local currency Augmented Dickey Fuller (ADF) test, H0: unit root, H1: no unit root (stationary) Phillips Peron (PP), H0: unit root, H1: no unit root (stationary) Kwiatkowski, Phillips, Schmidt and Shin (KPSS), H0: no unit root (stationary), H1: unit root Even at the 0.01 level of significance, the MVR tests generally suggest that many stocks in our dataset do not follow a random walk process. While the percentage of stocks significant for at least one of the q levels is substantially higher for the post-accession dataset than the pre-accession dataset, the z-tests reveal that this is not statistically significant. At the 0.1 level of significance, more than half of all stocks do not conform to a random walk process for at least one of the sampling intervals applied, and the results are very similar for the pre- and post-accession nations. Excluding Czech Republic, Latvia and Slovakia (the small number of stocks listed in these nations makes inferences about them questionable anyway), there is not a large variation amongst the different countries in our dataset, with the results for Poland and the entire region being almost identical. 4.4 Liquidity Controls Table 5 shows the results from the liquidity controls employed: The results from using market capitalization as a proxy for liquidity are shown in Table 5 Panel A. For both the pre- and post-accession datasets, smaller capitalized stocks exhibit higher levels of serial correlation. Runs tests are also substantially affected by their market capitalization quintile, with the smaller market capitalization quintile stocks returning a higher proportion of significant results. The ADF and PP tests are both excluded from the table as every stock in our dataset can be considered statistically significant at the 0.01 level and thus there is no variation across any of the liquidity quintiles. For the KPSS tests, the results for the large market capitalization quintile are very similar to those from the small market capitalization quintile, therefore there is nothing to suggest that the KPSS tests is affected by liquidity (as measured by market capitalization). For the MVR tests, portfolio 5 actually has a higher percentage of stocks returning statistically significant results than any of the other four quintiles: lack of liquidity is clearly not distorting results from the MVR tests. Whilst lack of liquidity associated with smaller market capitalization may have distorted some of the tests of serial independence, a substantial number of stocks in the largest market capitalization portfolio still return significant results. Market capitalization does not have any meaningful effect on any of the three unit root tests of the MVR tests. The results from using average volume divided by shares outstanding as a liquidity control are shown in Table 5 Panel B. For serial correlation, the number of stocks significant at each of the three significance levels we use is actually higher in the most liquid portfolio 5 than in the least liquid portfolio 1. Therefore, there is no indication that lack of liquidity, as measured by average volume divided by shares outstanding, is distorting the serial correlation tests. Whilst the runs tests return the highest percentage of significant results for the lowest-liquidity portfolio 1, there is not a huge amount of variation across the quintiles. In a similar manner to the serial correlation statistic, the percentage of stocks returning significant results for the KPSS tests actually increases as liquidity increases. The MVR tests return very similar results across the five quintiles. It is clear that Table 5 Panel A: Liquidity Controls - Market Cap v lp o e t h n t a c o e in n gi o t s s a el c o t t a S sl 8 ™ S v0 r2 e- t4 n0 i0 g2 ni fS £ o a2 " o 0 0 2 st80 sTe00 t s e n4 u0 R0 2-0 0 0 2 8 0 s0 un20-4 uf R40 rf o020 |S £ o u2-z o 0 0 2 iS s o C 555 0 0, ,9 Lnt— v o. (5 O jz c ro u O 4= > ^ cu " O -S £ « « äS ^ J2 5gs=c§ ^ U — (Ü(N OX^ò -= +j(D 2 3 m O O o oc -o C ^ il ^ (U -O « O O Ooöö c ^ oo 3 -Q a o C U 3 -Q a o a> ru > £ <ü — LO 3 -D a o C C 5 19 24 0 0 0 Total 65 24 89 65 24 89 Correct 60 19 79 65 0 65 % Correct 92.31 79.17 88.76 100.00 0.00 73.03 % Incorrect 7.69 20.83 11.24 0.00 100.00 26.97 Total Gain* -7.69 79.17 15.73 Percent Gain** NA 79.17 58.33 Estimated Equation Constant Probability Dep=0 Dep=1_Total_Dep=0 Dep=1_Total E(# of Dep=0) 58.26 6.74 65.00 47.47 17.53 65.00 E(# of Dep=1) 6.74 17.26 24.00 17.53 6.47 24.00 Total 65.00 24.00 89.00 65.00 24.00 89.00 Correct 58.26 17.26 75.52 47.47 6.47 53.94 % Correct 89.63 71.91 84.85 73.03 26.97 60.61 % Incorrect 10.37 28.09 15.15 26.97 73.03 39.39 Total Gain* 16.59 44.94 24.24 Percent Gain** 61.53 61.53 61.53 *Change in "% Correct" from default (constant probability) specification **Percent of incorrect (default) prediction corrected by equation Source: Author's calculations in EViews 6 Results of the Hosmer-Lemeshow test and the Andrews test are presented in the following table. A high value of the Andrews goodness-of-fit test and a low level of the Hosmer-Lemeshow test are desirable. Considering the Hosmer-Lemeshow test, if the associated p-value is significant (p<0.05), it might be an indication that the model doesn't fit the data. Since the H-L goodness-of-fit test statistic is much greater than 0.05, the null hypothesis that there is no difference between the observed and model-predicted values of the dependent variable is not rejected, implying that the model's estimates fit the data at an acceptable level. The next graph represents the forecasted probability of systemic banking crisis calculated from the dynamic logit model. The model sends signals within the signal horizon that is defined 24 months preceding the crisis - from November 2006 to October 2008. As it can be concluded from the graph, the highest probability of systemic banking crisis is during the first year of the signal horizon. This suggests that the model sends warning signals in the early stage, namely a year before the beginning of the crisis. Table 5: Results of the Hosmer-Lemeshow test and the Andrews test Quantile of Risk Dep =0 Dep: =1 Total H-L Low High Actual Expect Actual Expect Obs Value 1 8.E-07 0.0002 8 7.99958 0 0.00042 8 0.00042 2 0.0004 0.0013 9 8.99276 0 0.00724 9 0.00725 3 0.0014 0.0043 9 8.97551 0 0.02449 9 0.02456 4 0.0049 0.0106 9 8.93227 0 0.06773 9 0.06824 5 0.0160 0.0423 9 8.77340 0 0.22660 9 0.23245 6 0.0479 0.1375 8 8.19409 1 0.80591 9 0.05134 7 0.1727 0.3344 7 6.71711 2 2.28289 9 0.04697 8 0.4049 0.6251 3 4.29662 6 4.70338 9 0.74874 9 0.6448 0.8755 2 1.80876 7 7.19124 9 0.02531 10 0.8926 0.9997 1 0.30990 8 8.69010 9 1.59156 Total 65 65.0000 24 24.0000 89 2.79683 H-L Statistic 2.7968 Prob. Chi-Sq(8) 0.9465 Andrews Statistic 42.1494 Prob. Chi-Sq(10) 0.0000 Source: Author's calculations in EViews 6 Graph 1: The forecasted prob ability of systemic banking crisis 1.00.80.60.40.20.02005 2006 2007 2008 2009 2010 2011 2012 I — crisisfI A \l 1 A ! 1 iL VÄL Source: Author's calculations in EViews 6 In orderto check the robustness ofobtainedresults, Bayesian modelaveragingis als o apphc el As Bc°e aky et al. Ì2012, p. ebeaae suggeot,tha following hneas r^j^ressehnmo(^el sho uld be considered: y= a + X ß + £ l r yry £ ~ (0, S2I) wherey is a dummy variable denoting crisis, ay is a constant, ßy is a vector of coefficients, and £ is a white noise error term. Xy represents a subset of all available relevant explanatory variables, i.e. potential early warning indicators X. The number K of potential explanatory variables yields 2K potential models. Mark y is used to refer to one specific model from 2K models. The information contained in models is then averaged using the posterior model probabilities that are considered under the Bayes' theorem: p (My, | y, X) - p ( y | My, X) p (My) where p(My, | y, X) represents posterior model probability, which is proportional to the marginal likelihood of the model p (y | M , X) times the prior probability of the model p (M,). The essence of Bayesian model averaging is assigning weights to estimated models in order to determine which models have the best performance. For this purpose it is necessary to calculate the Schwarz Information Criterion as one of the most commonly used information criteria in order to determine which specification is more appropriate for the data nature. This criterion is known as the Bayes Information Criterion which is actually approximation of the Bayes Factor. A higher value of weight is given to the model with a smaller value of SIC, thus the model that has a smaller value of SIC is considered to be a more favorable specification. As already mentioned, using logit regression it is not possible to rank indicators according to their relative prognostic power when predicting systemic banking crises. This disadvantage can be partially overcome using Bayesian model averaging, because it is possible, by applying this technique, to assign adequate weights to simple logit models with at most two explanatory variables. Although individual variables do not have weights, their relative importance can be approximately determined on the basis of weights assigned to the model that contains these variables. Estimation results of implementation of the Bayesian model averaging technique are presented in the following table. On the basis of weights assigned to individual models that are calculated using SIC, it may be concluded that estimated models have very similar performances. The best performance is that of the model with explanatory variables Monex20 which represents one of two indices on the Montenegrin stock exchange and net loans with weight 0.16408. The model with the lowest performances is one that contains variables - 3-month Euri-bor and monthly growth rate of consumer prices with weight 0.12907. Marginal effects of explanatory variables are presented in the following table. Table 6: Estimation results of implementation of the Bayesian model averaging technique Model Variable Coefficient Statistic significance Weight (0-1) ASSETS 106.23 0.0001 Model 1 DEPOSITS -69.62 0.0010 0.14370 CAPITAL 13.42 0.0153 Model 2 BORROWINGS 19.33 0.0003 0.13973 LOANS 50.23 0.0000 Model 3 RESERVE_REQ -11.66 0.0205 0.15971 EURIBOR_1M 5.35 0.0043 Model 4 LLP 16.08 0.0024 0.13106 LOANS_DEPOSITS 37.15 0.0010 Model 5 INT_INCOME 7.60 0.0226 0.13266 EURIBOR_3M 6.06 0.0138 Model 6 PRICES_M 1.44 0.0113 0.12907 MONEX20 -9.46 0.0011 Model 7 NET_LOANS 47.32 0.0000 0.16408 Source: Author's calculations in EViews 6 Table 7: Marginal effects Variable Marginal effects ASSETS 16.28 DEPOSITS -10.67 CAPITAL 2.22 BORROWINGS 3.19 LOANS 7.46 RESERVE_REQ -1.73 EURIBOR_1M 0.80 LLP 2.41 LOANS_DEPOSITS 5.87 INT_INCOME 1.20 EURIBOR_3M 0.98 PRICES_M 0.23 MONEX20 -1.25 NET_LOANS 6.24 Source: Author's calculations in EViews 6 Application of the Bayesian model averaging technique represents an important part of the analysis. Namely, this technique enables estimation of more variables that can be relevant indicators of systemic banking crises, than it would be possible by using only a regular logit model. Putting a higher number of variables in one single regression may cause problems, such as multicolinearity. As can be seen, the dynamic model has captured eight variables, while using the Bayesian model averaging technique 14 variables are included where six of them are the same as in the dynamic logit model. Instead of estimating only a set of simple logit regressions, Bayesian model averaging gives an insight into relative importance of some variables in comparison with other variables. Therefore, it is possible to determine which indicators are more reliable for prediction of systemic banking crises. 4 INTERPRETATION AND DISCUSSION McFadden R2 indicates a relatively good goodness-of-fit of the estimated model. Results of the estimated dynamic logit model suggest that loans have the highest marginal effect on the dependent variable. Therefore, if this indicator increases by 1%, the estimated probability of occurrence of the systemic banking crisis will increase by 3.10, holding constant the remaining variables. If the value of variable LLP that represents loan loss provisions increases by 1%, the probability of systemic banking crisis will increase by 1.50. Also, if the loans-to-deposits coefficient increases by 1%, the probability of systemic banking crisis will go up by 0.02. On the other hand, if deposits increase by 1%, the probability of systemic banking crisis will decrease by 2.14. If capital increases by 1%, the probability of systemic banking crisis will increase by 1.26. Considering macroeconomic variables, it can be concluded that if 1-month Euribor increases by 1%, the probability of systemic banking crisis will go up by 0.35. Similarly, if EUR/USD exchange rate increases by 1%, the estimated probability of occurrence of systemic banking crisis will decrease by 1.09. Montenegro is a euroised economy, and one of the main advantages of fixed exchange rate regimes is that they enable achieving the macroeconomic stability thanks to a solid nominal anchor. However, it is necessary to emphasize that fixed exchange rates do not a priori provide macroeconomic stability. The main deficiency of fixed exchange rates is that they reduce flexibility of monetary policy. The reason for considering EUR/USD exchange rate as an early warning indicator is that Montenegro is a small and open euroised economy, so the trend of this variable might have a significant impact on the domestic economy. Concerning inflation, if the annual growth rate of consumer prices in Montenegro increases by 1%, the probability of systemic banking crisis will increase by 0.06. One of the most important variables that are related to international indicators is economic growth of the country that represents the main trading partner of the domestic country. According to available data starting from 2005, the largest portion of Montenegro's trading exchange, taking into account both export and import, has been realized with Serbia, therefore the most significant trading partner of Montenegro is Serbia. If the index of industrial production in Serbia increases by 1%, the probability of systemic banking crisis occurrence will decrease by 0.005. It can be concluded that it is a variable with the lowest marginal effect in this model. Seven simple logit regressions that individually have two explanatory variables are estimated, and thus there are 14 statistically significant indicators, while in the previous dynamic logit regression there are 9 indicators. Adequate weights have been assigned to all seven regressions using the technique of Bayesian model averaging. These results largely coincide with results of the previously estimated logit model. If indicator that represents total assets in the banking system increases by 1%, the probability of systemic banking crisis occurrence will increase by 16.28, holding constant the remaining variables. Similarly, if loans increase by 1%, the probability of systemic banking crisis occurrence will go up by 7.46, and if net loans increase by 1%, the probability of systemic banking crisis occurrence will increase by 6.24. If loan loss provisions increase by 1%, the probability of systemic banking crisis occurrence will go up by 2.41. That can be explained by the fact that banks approved more risky loans during credit expansion, therefore, relatively shortly after that, they had to allocate a larger amount of loan loss provisions. If deposits increase by 1%, the probability of systemic banking crisis occurrence will decrease by 10.67. Also, if the loans-to-deposits coefficient increases by 1%, the estimated probability that the systemic banking crisis will occur increases by 5.87. If capital increases by 1%, the probability of systemic banking crisis occurrence will go up by 2.22. Also, if borrowings which banks mostly take from their parent bank increase by 1%, the probability of systemic banking crisis occurrence will increase by 3.19. Variable reserve requirements represent one of very few monetary instruments which the Central Bank of Montenegro has at its disposal, since Montenegro is a euroized economy. Actually, it is more appropriate to say that it is a liquidity instrument. If this variable increases by 1%, the probability of systemic banking crisis occurrence will decrease by 1.73. If 1-month Euribor increases by 1%, the estimated probability that systemic banking crisis will occur increases by 0.80, and with the increase of 3-month Euribor by 1%, the probability of systemic banking crisis occurrence will increase by 0.98. If interest income increases by 1%, the probability of systemic banking crisis occurrence will go up by 1.20. Also, if the monthly growth rate of consumer prices in Montenegro increases by 1%, the probability of systemic banking crisis occurrence will increase by 0.23. Finally, if variable Monex20 increases by 1%, the probability of systemic banking crisis occurrence will decrease by 1.25. Indicators relating to a credit boom thanks to very good performances, have a dominant role in early warning models for systemic banking crises. The accelerated economic growth influenced the banks to initiate the exaggerated lending activity that led to credit expansion with three-digit yearly credit growth rates, and that in turn even additionally encouraged overheating of the economy. Funds taken as borrowings from parent banks during the credit expansion were mostly used for the lending activity. It was just a question of time when it would come to the bursting of the bubble that reached enormous proportions especially on the housing market. Besides developments in the domestic banking sector and in the overall economy, the crisis occurrence is also accelerated by negative global trends influenced by the global economic crisis. It is interesting that some indicators related to macroeconomic developments in the region and in the European Union, have also shown very good performances. These are 1-month Euribor and 3-month Euribor, EUR/USD exchange rate and the index of industrial production in Serbia. Therefore, it can be concluded that the Montenegrin economy and the banking system are exposed significantly to the trends on the global level. Developments on international markets have a significant impact on the domestic banking system and its stability, and therefore on the probability of systemic banking crisis occurrence. 5 concluding remarks Although many economists, especially critics of economics as science, consider that these models have proved to be unsuccessful because they failed to predict occurrence of the present global crisis, the economic policy can not be conducted today in an appropriate and efficient manner without reliable quantitative information. However, it is necessary to take into account qualitative estimates made by economic experts. The use of early warning models for systemic banking crises have to be adequately integrated within broader analyses that take into consideration all important aspects, as it is inevitable that some of these aspects will be overlooked by one of these models. These models can have an important complementary role as an objective measure of the banking system vulnerability. Regarding developing countries, it should be taken into account that they usually go through the catching-up phase in order to reach developed economies, and therefore they have higher economic growth rates. Economic growth during that phase is relying largely on the lending activity and it is sometimes difficult to differentiate between the credit expansion and the increased credit activity. Results of the estimated models have shown that the systemic banking crisis in Montenegro has its roots in the domestic economy. Causes of crises originate from the period of unsustainable credit expansion. A very low level of credit activity during the period before the beginning of the credit expansion has encouraged banks to race for a market share. Also, results have shown that although roots of crisis are in the domestic economy, there is a significant impact of international trends on the Montenegrin banking system and overall economy. Acknowledgements I would like to thank professor Jesus Crespo Cuaresma for his very helpful comments. The views expressed in this paper are those of the author and do not necessarily represent the position of the Central Bank of Montenegro. references Babecky, J. et al. (2012). Banking, Debt, and Currency Crises: Early Warning Indicators for Developed Countries. European Central Bank Working Paper, 1485. Crespo Cuaresma, J. & Slacik, T. (2009). On the determinants of the currency crisis: The role of model uncertainty. Journal of Macroeconomics, 31. Demirgü^-Kunt, A. & Detragiache, E. (1998). Financial Liberalization and Financial Fragility. IMF Working Paper, 83. Eichengreen, B. & Rose, A. (1998). Staying Afloat When the Wind Shifts: External Factors and Emerging-Market Banking Crises. NBER Working Paper, 6370. Gujarati, D. N. (2004). Basic Econometrics. Fourth edition. New York: McGraw-Hill. Kaminsky, G. L. & Reinhart, C. M. (1996). The Twin Crises: The Causes of Banking and Balance-of-Payments Problems. International Finance Discussion Papers, 544. Kaminsky, G. L., Lizondo, S. & Reinhart, C. M. (1998). Leading Indicators of Currency Crises. IMF Stuff Papers, 1. Wooldridge, J. M. (2002). Introductory Econometrics: A Modern Approach (2nd ed.). Southwestern Educational Publishing. SHAREHOLDERS' PAY-OUT-RELATED THRESHOLDS AND EARNINGS MANAGEMENT JERNEJ koren1 ALjošA valentinčič2 Received: 24 August 2013 Accepted: 5 September 2013 ABSTRACT: We investigate the thresholds in net shareholder pay-outs (dividends, share buy-backs and issuances) of a large sample of UK quoted firms. Discretionary accruals are analysed at these thresholds in relation to earnings management. By examining distributions and using a robust test for discontinuities, we show the existence of thresholds at zero bins of variables in question. Additionally, by looking at differences in means and medians of discretionary accruals in sorted distributions, we find that they are statistically different from bin to bin in vicinity of previously identified thresholds. 1. introduction and prior research Earnings, as the primary performance indicator of a firm, can be managed with the intent of companies reaching expectations-set performance thresholds (Burgstahler & Dichev, 1997) meeting analyst forecasts (Degeorge, Patel, & Zeckhauser, 1999), satisfying certain contractual obligations or fulfilling liabilities stemming from borrowing activities. Earnings management is also observed around certain corporate events, for example stock offerings or acquisitions (Erickson & Wang, 1999) or in connection with managers' compensations and bonus schemes (Bergstresser & Philippon, 2006). Still, earnings management cannot only be seen in a negative light. Under certain conditions it may also be beneficial for owners - through application of manager's acquired expertise in forecasting earnings or not dismissing a hired manager (who is good-working) too fast (Arya, Glover, & Sunder, 1998) - or at least neutral in a way that decisions taken with managed earnings in consideration are the same as they would be had earnings not been managed (Ronen & Yaari, 2010). Among factors, assuming managers' threshold reasoning and, consequently, the possible appearance of earnings management, is also a company's dividend policy. Dividend policy is determined by the company's management and, as there is no unique rule about the dividend policy, similarly efficient and successful companies can - and do - have different dividend pay-outs (Brigham & Ehrhardt, 2005). Miller & Modigliani (1961) proposed 1 University of Ljubljana, Faculty of Economics, Slovenia, Ljubljana, e-mail: jernej.koren@ef.uni-lj.si 2 University of Ljubljana, Faculty of Economics, Slovenia, Ljubljana, e-mail: aljosa.valentincic@ef.uni-lj.si a model of dividend irrelevance where corporate value should not be related to pay-out policy in a perfect and frictionless capital market.3 Excluding taxes and transaction costs, investors should thus be indifferent between (cash) dividends and capital gains. Historically, this has not been the case. Lintner's (1956) first study of dividend policy found that managers are reluctant to cut dividends and are willing to increase them only gradually after they are convinced of enough support of a higher level of dividends in the form of higher future earnings. Existing dividend levels thus act as a strong benchmark. In seeking to explain investor preferences for (cash) dividends, Shefrin & Statman (1984) put forward two explanations. Firstly, one of "self-control" where investors decide to consume only from dividends, not portfolio capital and are thus demanding dividends. Secondly, following Kahneman & Tversky's (1979) behaviour theory proposition that losses loom larger than gains, dividends are preferred by people who are averse to regret (a potential increase in share price had they sold their stock instead of receiving a dividend). The behaviourist view can also be a potential explanation for dividend decreases having a more negative market effect than dividend increases. If dividends and their levels present a benchmark for investors, market reactions to dividend changes, especially downward, are found to be substantial (e.g., Grullon, Michaely, & Swaminathan, 2002). Bhattacharyya (2007, pp. 9-10), for example, also provides a short overview of stylized facts on dividends. A company's dividend policy can be affected by various factors such as market imperfections, behavioural considerations, firm characteristics or managerial preferences (Baker, Powell, & Veit, 2002). They differ in importance to individual firms, but they form the basis for possible earnings management. While the latter two factors include firm- and management-specific factors, the former two factors comprise broader aspects such as different tax treatment of dividends and capital gains, overcoming information asymmetries with signalling new or additional information and shareholder and investor clienteles that favour dividends in various degrees at various times (see Baker & Wurgler (2004) for a catering theory of dividends). The distributional analysis and existence of thresholds was first suggested by Hayn (1995) who points out the discontinuity of earnings around zero in her study of the information content of loses.4 Building on this empirical irregularity, Burgstahler & Dichev (1997) show that firms manage earnings to avoid reporting loses or earnings decreases. They interpret low frequencies of small loss (earnings decline) observations and high frequencies of small profit (earnings increase) observations as a consequence of firms' active efforts to cross the loss (earnings decline) threshold what results in a migration of observations to the right of such divide as seen if a distribution is plotted. Assuming that without 3 DeAngelo & DeAngelo (2006) contested that pay-out policy is not irrelevant as put forward by Miller & Modigliani (1961) but their proposition was reconciled as having assumed different agency costs (Handley, 2008). 4 An interesting case of goal reaching behaviour research is also the analysis of Carslaw (1988) who finds abnormal distribution of income numbers in financial statements with the bias tilting towards numbers just above multiples of powers of ten (i.e., N X 10k) as cognitive reference points. earnings management the distribution of earnings would be fairly smooth, they test the documented asymmetry around zero (earnings or changes in earnings) thresholds. Their findings are confirmed by Degeorge, Patel, & Zeckhauser (1999) who add another threshold of meeting analyst forecasts (i.e., avoid earnings surprises). Additionally, they establish a hierarchical order of the three with positive earnings threshold being predominant, followed by not falling short of previous earnings and lastly meeting analyst expectations. Critique of distribution analysis is based mainly on the effect of deflator and the sample selection procedure, both of which can have an effect on the resulting distribution (Durtschi & Easton, 2005). If the deflator differs systematically between profit and loss firms it can move the scaled observations towards or away from zero, what is most commonly the case when scaling by market price, but also found for other deflators (Durtschi & Easton, 2009). Alternative explanations of the discontinuity include asymmetric effects of taxes and special items that also contribute to observed shapes of distributions (Beaver, McNichols, & Nelson, 2007). We therefore study thresholds and earnings management from the standpoint of attaining (expected) pay-outs to investors as earnings levels are often directly or at least indirectly connected to the pay-outs, e.g., in companies with fixed pay-out ratio policy or linked to various contractual obligations that set limits on pay-out possibilities. The first study in this area is the analysis of Finnish companies that managed earnings to ensure constant dividend pay-out to large institutional investors who prefer stable dividends (Kasanen, Kinnunen, & Niksanen, 1996), whereas, in the US, Daniel, Denis & Naveen (2008) have shown that firms manage earnings upward to reach expected levels of dividends (defined as last year's dividend) when they expect they would otherwise fall short of it, proving they are important thresholds for managers. Similar findings are reported by Atieh & Hussain (2012) for UK. They show that earnings may be managed by firms which also try to avoid a decrease or even elimination of dividends and show a concern for coverage ratios, but the pressure is lower for larger firms which face less restrictive debt covenants. Debt covenants can impose restrictions on dividend payments if the financial position of the firm does not appear adequate. Moir & Sudarsanam (2007) report three quarters of firms in their study to have covenants attached to debt contracts. Another recent study by Bennet & Bradbury (2007) proposes dividend cover to be considered as a threshold as firms are likely to manage earnings to avoid cutting dividends, i.e., keeping them at least at their prior year's values. A comprehensive survey of CFOs by Brav et al. (2005) shows that managers are willing to go great lengths to avoid a dividend cut but increases in dividends are a second-order concern. The authors also observe that share repurchases have become an established alternative pay-out instrument to dividends. However, they do not convey the same signals about companies' future behaviour or performance. Dividends are seen as a more permanent commitment to provide shareholders with a reasonably stable cash flow, whereas repurchases (particularly the ones on a discretionary and non-constant basis) are viewed as more flexible. Repurchases would now be the primary choice of many firms had their dividend history not existed. Interestingly, little support is found for the signal- ling hypotheses, that is, not many managers state they are paying dividends to convey a company's true state (future prospects) or to intentionally separate them from competitors. Taxes are also not a primary concern in deciding about the payment/increase of dividend or in choosing between them and repurchases. Repurchases are gradually replacing dividends as the primary pay-out method with higher correlation to possible swings in earnings levels (with a shorter lag than for dividends). Skinner (2008) reports that firms which pay dividends only practically do not exist anymore. Other research has also found a decline in dividends paid by US listed firms, attributing it to both different firm characteristics and lower propensity to pay in general (Fama & French, 2001). Contrary to the latter, Grullon & Michaely (2002) find repurchases to be important in substituting dividends and US corporations financing them with funds that would have been otherwise used for dividend increases. What further motivates our research is a finding of Hribar, Jenkins & Johnson (2006) who assert that share repurchases are used by some companies to reach analysts' earnings per share forecasts. This implies that repurchases might be viewed as a possible earnings management tool. In this paper we analyse a UK sample with focus on three theoretically possible shareholder-related cash flows.5 Next to dividend pay-outs we also consider share repurchases and issuances of new shares, where the company is receiving funds from investors, resulting in a "negative" pay-out to shareholders. As these three shareholder-related cash flows might all be broadly regarded as dividend (pay-out) related decisions, we investigate the existence of thresholds in all three cases. This view is in line with Ohlson's (1995) valuation model that confirms Miller & Modigliani (1961) value displacement property as dividends are paid out of book value and consequently reduce market value on an one-for-one basis rendering dividend policy irrelevant. Ohlson's model allows (requires) negative net dividends, i.e., capital contributions (share issuances) exceeding pay-outs. As accruals, and more precisely their discretionary component, are often associated with lower earnings quality and possible earnings management, (e.g., see Dechow, Ge & Schrand (2010) for an overview) we are also interested to what extent discretionary accruals are present at the hypothesized pay-out thresholds. Although Yong & Miao (2011) find that dividend paying status is associated with the quality of earnings in general, they also find that the association is stronger when dividends increase in size. Therefore, inspecting the margin of dividend payment or dividend increase would be informative since firms potentially having difficulties in reaching these thresholds could still make use of discretionary accruals to arrive to them. H1: Companies attempt to reach thresholds of net shareholders pay-outs, which results in breaks in distributions of net shareholder pay-outs. H2: Thresholds are associated with significant discretionary accruals levels. 5 Beginning of section 3 (Sample selection and description) explains our choice of the UK market. This study helps to determine if repurchases and new share issuances, although not typically regular events, affect the pay-out level targeted by the management. This would broaden the perception of flows that are viewed as important in setting companies' dividend policy. In the process, a robust test of discontinuity of distribution is used (Garrod, Ratej Pirkovic, & Valentincic, 2006). Moreover, discretionary accruals as a proxy for earnings management are analysed in relation to the pay-out levels. The remainder of the paper is structured as follows. Section 2 presents the research design employed in our analysis, followed by sample selection and data description in the next section. Section 4 presents main empirical results and section 5 reports additional tests. Section 6 concludes. 2. RESEARCH DESIGN We begin our investigation by constructing the variables representative of pay-out-related thresholds. Typically, dividend pay-outs are investigated, either in their total amount or as change from year to year, both relative to opening total assets to account for size differences among firms. We denote DIV as the ratio of dividends to lagged total assets and D_DIV as the ratio of change in dividends from the previous year to the current year, scaled by lagged total assets. The variable D_DIV_DIV scales the dividend change from the previous year to the current year by previous year's dividends level to get a variable representing relative yearly pay-out changes. We calculate net shareholder cash flows as the sum of all cash flows investors might be dealing with, i.e., dividends received plus stock repurchases (as positive cash flows from the company to shareholders) less any share issuances in a given year (negative cash flows from shareholders to the company): net shareholder cash flows = dividends + share repurchases - share isuances Analogous to the dividend variables above, NSCF denotes the ratio of net shareholder cash flows to lagged total assets, D_NSCF scales yearly changes in net shareholder cash flows by lagged total assets and D_NSCF_NSCF is the change in net shareholder cash flows scaled by its lagged value. We also calculate and perform initial analyses on the scaled sums of dividends and stock repurchases only but, as dividends are highly dominating this sum, the results do not differ in any important way from dividend-only findings and offer no incremental insights. This part is therefore not investigated further in this paper. Variables as defined above are then distributed into bins of widths 0.005 for total assets scaling and 0.01 for pay-out scaling.6 That corresponds to forming groups that contain 6 These bin widths were selected for both, comparability with prior research investigating distributions, although of different analysed and scaling items (Burgstahler & Dichev, 1997; Durtschi & Easton, 2005; Bennet & Bradbury, 2007) and ease of interpretation. As setting the bin width can have a huge effect on the histogram 156 ecoNoMIC and BUsinESS reviewi VOL.15 |No. 2| ^013 c^tisjjr-vattions waith vahier weit lein 0 . 5% of iacoen total assets cr 1% of tanned jEay-oot. IION5ce inoremsnte ore oIso f sed lor alf jantas^tjutnO bins. Bios widfhw Nor f-y-oet scaling ate Iwlce rs bis ac for OwHtaC aocete seating because thu lotCer oea nrirasi kt-cr irr cfsolutevalbr awosl thecr uoe ac a d^noanlria^ar iteset 2ts 1 n natikala cmantr r;irfoa tha2 have 5r be prtsenteti with nighgE acgiragy tn yiao'vreiQ^aräfic^lai"^istoc^^^sn(0ver-smoothing" (Scott, 1979). All bins are defined to include lower bound as we want the central bin to include observations with zero onil staili poshivn vrOuer and NxtltdN the souor bound. Ahhoueh cailc^S^tr1E, ws constder zero (ac COe scaled amarro. or chcdge, whree apflirsblr) as c1ee Iììmn^ noctf negativa rignalling vahrn rwC thos alte iluoe^^olsI at- mvastigete. Tlse go -allied Nrecn 1c1 n" as eheonfors definad ae innluaing x i f0^eeloLgo O.gM usmg FreedmantDlaconis1 Sermula (h = ) to oee r 107 uiing (hr gtuago rs' eu^e (h also dependent on the variable. The latter widths were drastically reduced to 0.400 or less if outliers at f% were removed before the calculation. Suggested bin widths obtained using the Scott's formula (h = 3.49 X cn^'^werebetweenthetwo extremes. Hn; GRPV tEE. sEaltEitsic; t, derived from ^ebyElscvinequaPiyiscomputed as foUowsm ecuatkn (1 btlow, whüe Essummo; mCependenS evErts hHate uHtv^^ cnpucs ^C1) =N x =. and (X) = Nxp.p (l - /u), whert Nis fhe Eotal nututar oi1 ibrnisEationsmAe sosnple and p.is the xrvbabihtythat: mobs-rns-ion wüSfaHinto mternam^primaril^Homputed aj en aveuagu ot1 Cwo sdjpRvn( interesla: p. = xi-t~xi+t. 2N X - E(Xi) Ti=7^XŽT (1) wdencXisChl -c t-rlnumber rf oVEernetivnsln 1 nCsrval (i). Values of the test statistic are tubulaled in Gmtpcd, Ratej diEkeuk D V-knlmciv (,2006) and a breakin She distribution at inteEual ils iPeniifkd aTstandordsignificcnce luveln odT0%, 5% and 1% corresponding So abaolute vaiola oEThs t slcCtetiKc of3.16, 4.47 and10.00 respectively. OREe tee etea in) eEisteU In the soIe amru als hneee at SURl hypot hasizedthr eiAa lde and in -arderolsi atue LisCTi^^on^t.71 com]3oneni oi1 tnolal schERlls. We uoekhe modified Jonvrmodd JDeellon, Sìtph, & Sweemy S095C He eelimltenon-clievlelioneryeccru-oCf wh)ch we °hEn use ta Astamius Rhe dinereHiAn-En lompHn^nt oL aLeluols as the diffcnen-e between eetìmviLi) valsei ant toi accruolv. Firstlyiotal accrnaco lrerom-paiiieO as: TACC c(ACAt - ACLt - ACas/it -I- ASTDt w Deot)dT-t-i (2) seNseae A(LA(is thn chants ln nunevi Lsseie( ACLu1n iCechange lis ^u^re'enl^iaACfi^i^s, ACa sh (a (Eie ahenge itc cavf ant ccdt equini: n7s( kSTD t ls tlie rlaanise tsi vhorT te r m d sD;tJ D ep tre dtprTciation aod amortivltion chargas andOlS^ art lagg;sdl^olal ast c.^.Tlie modifi ed Joneo modelisof She following form: NDACC = nl(l/nAt_l) + n2(AREVt - ARECt) + n3(PPEt) (3) assets, AREVt ia^ho^lt^g^ intevenues,sc ale dbylag gcdtr tal assets, Cl jUC^s thechangelnreceinpbks , ^caleOlg lagetdtslal asnJs and i^g_E(rcgr oss proportyplanh and equipment, toalrg bylaggil totalarsetr. Estimates otcg, a2 and a3 areohtained t^^^^t^im^^réhg^li^naodelin equation ertl^^icitoslslenur^l^aiti^fol accsltals on tCrkft-hanll ^iu^^.Tt^^^otim^^^C^o^lS^^iet^t^^a^erP^n i^^oit tsur^^i^i^i^lrnp^-r^lats^tionary accruals. hliei^^^it^irgl^sg tlslr modelare dlscreléoharp acarhals.Dlscpetlonary ac^ua^antlien analysed bm-clr e forfosclbleO^^Ufet^nocr Shlhelsm r^ngrmWlgn mUtei.For Uhisp ur-g(ttrllhultJctJ formnansanPrUeWikoxsos a^^^-^m^orrr^^i^iat^^ grendf.We oxpect slotéat ieoilypioni fica rt^i^ff^i^^ i^cesa^Ol^^^^r^ tlomaiy acrimsesinbinsang undzero thre sh-olgdthatvguiC llnklti two potential indicators of earnings management and suggest discretionary accruals' use in connection with these thresholds. 3. sample selection and description We acquire data of publicly listed UK companies from Datastream. This market is selected because companies in the UK have historically paid considerable dividends that still persist. A large majority (almost 85%) of UK firms paid dividends in the 1990s and dividend pay-outs dominated proportion-wise, although repurchases have been on the rise (Renneboog & Trojanowski, 2005).7 Even recently, despite the trend of declining pure dividend pay-outs (Skinner, 2008), UK firms still tend to pay out dividends relatively more often than elsewhere (Denis & Osobov, 2008). As we want to have the initial sample as broad as possible, companies in the period from 1990 to 2012 are considered. Prior to 1990, the lack of data availability hinders a more detailed analysis and an incomplete set of companies' financial information was provided for 2012 at the time of data collection. A note is necessary about dividend inputs from the database. Since IFRS-compliant reporting became mandatory for all listed companies in the EU for annual periods beginning on or after 1st January 2005, a provision in the standards requires companies to account differently for dividends paid. Before 2005, under the Statement of standard accounting practice (SSAP 17 - Accounting for post balance sheet events, 1980), dividends were accounted for as an adjusting post balance sheet event in the period to which they related. After 2005, it is prohibited to recognise dividends declared after the end of reporting period as a liability to that same reporting period (IAS 10 - Events after the reporting period). Instead, such dividends are disclosed in the notes but accounted for in the period in which they are paid. Thus, actual pay-out liability has priority over its source (earnings). This results in reported dividends in period (t) consisting of final dividend for period (t-1) and interim dividend(s) for period (t). Final dividend for period (t) is then recognised in period (t+1) financial statements etc.8 Table 1: Sample construction procedure firm-year observations of listed UK companies in the period 1990 - 2012 38,429 observations with missing essential data 742 observations with zero total assets or sales 2,065 observations of financial and utility firms 5,380 final sample firm-year observations (3,177 distinct firms) 30,242 Notes: This table presents sample selection process. Starting sample of listed UK companies is obtained from Datastream and identified using nation code (WC06027). All financial industry related firms and utilities are excluded due to their specific operating properties. 7 Dividend payments have been more frequent in the UK also due to the more favourable tax treatment of dividends in the past (prior to the Finance Act 1997, see section Additional tests for more information) but remained high after the change as well. 8 For example, GlaxoSmithKline (GSK) in its 2005 annual report shows a breakup of dividends into four interims of (all in £m) 568, 567, 568 and 792 for 2005 respectively and 575, 573, 571 and 684 for 2004 respectively. But, since GSK normally pays a dividend two quarters after the quarter it is relating to, dividends actually paid in 2005 were the last two interims for 2004 and the first two interims for 2005. The sum of those, £m 2390, is then reported as dividends for 2005 and also found as the database entry. We first eliminate entries with missing data that are essential for the analysis, e.g., missing total assets or industry codes. We then remove observations with zero total assets and/or zero sales as these are not believed to be truly operational and the former would imply division by zero in the construction of our variables of interest. Lastly, as a common step, we remove firms from financial and utility sectors because of their operating specifics. We end up with 3,177 distinct companies and 30,242 firm-year observations as presented in Table 1. Out of these, 62% include dividend payments, 60% report proceeds from sale or issuance of stock and 11% show a change in redeemed, retired or converted stock. A substantial share of issuances indicates a possible large effect on NSCF, whereas the extent of repurchases is somewhat smaller than expected. Examination of the data also revealed some confounding entries in form of negative values of repurchases (14 identified) and negative values of issuances (134 such cases), both of them are not supposed to be negative following the definition of Datastream datatypes. A subset of each was, where possible, manually checked back against firms' annual reports and entries were corrected accordingly, e.g., into positive values. Lastly, otherwise sound observations with missing dividends, repurchase or issuance data had those set to zero.9 Table 2 presents sample structure by years. As there are no big deviations in any specific year, we can observe a first peak in the number of listed UK companies in 1997, followed by a slight decrease and another gradual but steady increase in the years following up to 2006. However, in the last years there is quite a strong decline coinciding with the development and deepening of the financial crisis. Data for 2012 were not fully populated at the time they were collected. Table 2: Year composition Year N in % Year N in % Year N in % 1990 1,154 3.82 1998 1,462 4.83 2006 1,551 5.13 1991 1,166 3.86 1999 1,389 4.59 2007 1,491 4.93 1992 1,147 3.79 2000 1,398 4.62 2008 1,352 4.47 1993 1,152 3.81 2001 1,443 4.77 2009 1,236 4.09 1994 1,184 3.92 2002 1,474 4.87 2010 1,124 3.72 1995 1,183 3.91 2003 1,502 4.97 2011 1,063 3.51 1996 1,467 4.85 2004 1,553 5.14 2012 658 2.18 1997 1,542 5.10 2005 1,551 5.13 Notes: Year distribution of the sample in presented in this table. Total number of observations is 30,242. At the time of data collection year 2012 was not fully populated, therefore the number of observations is accordingly smaller. Descriptive statistics in Table 3 suggests skewed distributions in almost all variables. As we are interested in the centre of distributions and especially in specific breaks, quar-tiles are reported along with the average, but standard deviations indicate that there are 9 There were 1,521 such cases, of those only 390 with missing dividends. Remaining missing repurchases and/ or issuances would prevent the construction of NSCF with dividends mostly available. Omission of these cases does not change the results. substantial extreme observations.10 The number of observations is mostly affected by the denominator, particularly when scaling by past dividends and less so when scaling by past NSCF. The first four variables use lagged total assets for scaling and are limited by that. Only DIV and D_DIV have comparable means and medians, dividends amounting on average to around 2% of previous year's total assets and dividend change being positive but of minor amount compared to total assets. The remaining four variables have means and medians differing in both sign and magnitude, once more implying skewed distributions. Table 3: Descriptive statistics Variable Mean 25% Median 75% SD N DIV 0.024 0 0.015 0.033 0.086 26813 NSCF -0.256 -0.001 0.009 0.030 6.274 26813 D_DIV 0.002 0 0 0.004 0.097 26813 D_NSCF -0.156 -0.012 0.001 0.016 9.047 26813 D_DIV_DIV 0.425 -0.004 0.084 0.241 11.836 17201 D_NSCF_NSCF 39.208 -1 -0.039 0.229 2,175 22829 Notes: This table presents descriptive statistics for analysed variables. DIV = dividends (WC04551) scaled by lagged total assets (WC02999); NSCF = (dividends + repurchases (WC04751) - issuances (WC04251)) = net shareholder cash flows scaled by lagged total assets; D_DIV = change in dividends from year (t-1) to (t) scaled by lagged total assets; D_NSCF = change in net shareholder cash flows from year (t-1) to (t) scaled by lagged total assets; D_DIV_DIV = change in dividends from year (t-1) to (t) scaled by lagged dividends; D_NSCF_NSCF = change in net shareholder cash flows from year (t-1) to (t) scaled by lagged net shareholder cash flows. Number of observations varies due to availability of respective denominators used in variables' construction. For visual representation, we plot histograms of respective variables sorted into bins 0.005 or 0.01 wide as described in the previous section to arrive at distributions of interest. Almost all distributions imply a threshold at the zero bin, firstly in amounts relative to total assets (attaining dividends or non-negative net shareholder cash flows). Panels A and C in Figure 1 show striking mode bins of small non-negative pay-outs and a comparison of the two panels suggests that dividends clearly dominate also in NSCF calculation. Although halved in size (10,419 observations in bin(0) for DIV and 5,047 observations in bin(0) for NSCF), the zero bin of the latter is still clearly outstanding from the remaining distribution. There are also changes, with observations shifted to bins left of zero due to effect of issuances, but the distribution to the right of zero is not much different compared to DIV. Bin(0) modes disappear when observations equalling exactly zero are excluded in panels B and D. What remains is a mode in some of the subsequent positive bins (around 2-3% of lagged total assets) for both DIV and NSCF. While the zero bin in DIV is not standing 10 We did not exclude any outliers since our central analysis is concerned with specific observations at the centre of respective distributions. As all our variables are ratios, outliers can arise due to disproportionate numerators and denominators in the span of one year. This may be related to one variable only. Therefore, by excluding outliers relating to one variable we could lose economically-sound observations in other variables. out in any way, the one in NSCF is missing almost 400 observations (estimated as the difference to the average of adjacent bins) for a smooth, normal-like distribution. This case could indicate that NSCF are not a threshold of their own, in a way that firms would target its combined value as a reference point for investors. Figure 1: Histograms of selected distributions Panel A: Dividends scaled by lagged total assets fro. r £ ffl Panel B: Dividends scaled by lagged total assets (without os) & Is fs Panel C: Net shareholder cash flows scaled by lagged total assets Panel D: Net shareholder cash flows scaled by lagged total assets (without os) Notes: This figure presents distributions of variables of interest. Panels A and B graph DIV, with and without zero observations, and panels C and D graph NSCF, with and without zero observations. DIV = dividends (WC04551) scaled by lagged total assets (WC02999) and NSCF = (dividends + repurchases (WC04751) - issuances (WC04251)) = net shareholder cash flows scaled by lagged total assets. Bin width is 0.005 with lower bound inclusion, i.e., "zero bin" includes x if 0 < x < 0.005, "bin one" includes x if 0.005 < x < 0.01 etc. As observations of zero in given variables have such an overwhelming effect on distributions, they are not reported in Figure 2 but they are still included in the analysis that follows. Findings of clearly modular bin(0) are confirmed for scaled changes in dividends (D_DIV, Panel A) and scaled changes in net shareholder cash flows (D_NSCF, Panel C) - even without observations equalling exactly zero. What is of interest is that, in case of D_NSCF, the bin with the second highest frequency is actually the first negative (and not positive, as more commonly expected) bin and this pattern is even repeated bin-wise as we move away from zero bin. The negative effect issuances have on D_NSCF outweighs the combined positive effect of dividends and repurchases in these cases. Figure 2: Histograms of selected distributions Panel A: Dividend changes scaled by lagged total assets (without os) Panel B: Dividend changes scaled by lagged dividends (without os) Panel C: Net shareholder cash flows changes scaled by lagged total assets (without os) Panel D: Net shareholder cash flows changes scaled by lagged net shareholder cash flows (without os) Notes: This figure presents distributions of variables of interest. Panel A graphs D_DIV, panel B graphs D_DIV_DIV, panel C graphs D_NSCF and panel D graphs D_NSCF_NSCF, all without zero observations. D_DIV = change in dividends (WC04551) from year (t-1) to (t) scaled by lagged total assets (WC02999); D_DIV_DIV = change in dividends from year (t-1) to (t) scaled by lagged dividends; D_NSCF = change in net shareholder cash flows (= dividends + repurchases (WC04751) - issuances (WC04251)) from year (t-1) to (t) scaled by lagged total assets; D_NSCF_NSCF = change in net shareholder cash flows from year (t-1) to (t) scaled by lagged net shareholder cash flows. For panels A and C bin width is 0.005 with lower bound inclusion, i.e., "zero bin" includes x if 0 < x < 0.005, "bin one" includes x if 0.005 < x < 0.01 etc., and for panels B and D bin width is 0.01 with lower bound inclusion, i.e., "zero bin" includes x if 0 < x < 0.01, "bin one" includes x if 0.01 < x < 0.02 etc. Lastly, looking at pay-out changes relative to their lagged values (D_DIV_DIV and D_ NSCF_NSCF, Panels B and D, respectfully), zero bin threshold mode remains obvious in dividend changes scaled by lagged dividends, but with a lot lesser difference to surrounding bins. In the case of D_NSCF_NSCF zero bin practically blends in the distribution and does not even seem to represent a threshold on the left (negative) side, the distribu- tion itself not displaying any noticeable breaks whatsoever. This is once more suggestive that no systematic threshold attaining behaviour can be observed with regard to net shareholder cash flows. Frequencies of dividend increases and net shareholder cash flows increases relative to their lagged values rise and/or remain high up to bins denoting growth in the order of 10% (note that bin width is 0.01 in these two cases as the denominators are considerably smaller than total assets used beforehand). Another interesting observation is bin(10) of D_DIV_DIV, denoting cases of dividend increase between 10% and 11% compared to previous year's dividends. The bin in question appears to jut out of the distribution and is also statistically evaluated in the next section. 4. RESULTS We attempt to formally confirm observations derived from histograms in the previous section with the use of GRPV discontinuity of distribution test. Table 4 reports values of the GRPV test applied for all cases inspected earlier (with and without zero observations) and fully confirms our assumptions. In all six cases of zero values of variables included, zero bin represents a discontinuity from the remaining distribution, inferences being done at P-values far below 1% (critical values of the test in absolute terms for significance levels of 10%, 5% and 1% are 3.16, 4.47 and 10, respectively). The discontinuity is stronger in dividend-related variables compared to NSCF-related ones, implying that repurchases and issuances lessen the break to some extent by moving some observations away from zero bin. Scaling by total assets results in stronger breaks than scaling by lagged values of pay-out, suggesting that the choice of scaling variable also plays an important role in distribution analysis as also suggested by previous research (Dechow, Richardson, & Tuna, 2003; Durtschi & Easton, 2005). On the other hand, in cases where zero values of variables are excluded from distributions, discontinuity is still statistically confirmed in four out of six cases. The H0 of continuity of distribution cannot be rejected in the first (DIV) and last case (D_NSCF_NSCF) as suggested and anticipated by the histograms in the preceding section, whereas other variables have results significant at the 1% level although test values are considerably lower than before the exclusion of zeros. A comparison of the four variables representing scaled changes in either dividends or net shareholder cash flows shows consistently larger breaks in dividends. We thus regard them as the driving factor for threshold existence. The fact that breaks are lessened with the inclusion of repurchases and new share issuances implies that these are not used with the intent of reaching a NSCF-related threshold, but rather for other purposes. Table 4: GRPV discontinuity of distribution test (1) (2) zero observations included without zero observations DIV 376.61 1.42 NSCF 115.39 -11.36 D_DIV 336.51 123.70 D_NSCF 124.24 62.61 D_DIV_DIV 73.02 21.16 D_NSCF_NSCF 39.97 1.95 Notes: Reported are T values of GRPV discontinuity of distribution test for zero bins of variables analysed. First column reports test statistics computed including observations of zero in selected variables, and second column reports test statistics computed without these observations. With Hgg: the distribution function is continuous, the values of T at standard levels of significance are: at 10% |t| = 3.16, at 5% |t| = 4.47, and at 1% |t| = 10. As the number of observations in adjacent bins is required by the test, in the first row (case of DIV) bins on the left of zero (negative bins) are empty (there are no negative dividends) and are as such affecting test statistic computation. DIV = dividends (WC04551) scaled by lagged total assets (WC02999); NSCF = (dividends + repurchases (WC04751) - issuances (WC04251)) = net shareholder cash flows scaled by lagged total assets; D_DIV = change in dividends from year (t-1) to (t) scaled by lagged total assets; D_NSCF = change in net shareholder cash flows from year (t-1) to (t) scaled by lagged total assets; D_DIV_DIV = change in dividends from year (t-1) to (t) scaled by lagged dividends; D_NSCF_NSCF = change in net shareholder cash flows from year (t-1) to (t) scaled by lagged net shareholder cash flows. We therefore confirm breaks at zero bins in the distributions of scaled pay-outs, which is indicative of existence of thresholds. The exclusion of zero observations has different meanings, depending on the variable in question. The DIV variable is specific, as it is bounded to the left of zero, i.e., there are no negative cash dividends. Zero observations in this case are firms that do not pay dividends at all. Therefore, their exclusion is justified as they obviously do not try to attain any pay-out threshold. The majority of dividend pay-outs are concentrated in the first ten bins, i.e., up to 5% of previous year's total assets. Nevertheless, we keep the analysis of DIV in both versions as a reference. Similarly, in NSCF, it is practically never the case that the three components would sum up to exactly zero, meaning that zero observations are those of zero values in all three components and these again are validly excluded.11 This is not as straightforward in scaled changes of dividends and net shareholder cash flows. D_DIV or D_DIV_DIV equal to zero may indicate a non-payer, but it can also indicate a no-change in dividends, keeping their level unchanged from the previous year. Analogously, D_NSCF and D_NSCF_NSCF values of zero can mean non-payers, no-changes in the sense that the firm only pays dividends and does not use repurchases and/or issuances or rare cases of the NSCF components summing exactly to zero. We also separately evaluate bin(10) of the D_DIV_DIV distribution. The value of the test statistics of the GRPV test amounts to 6.22 and is significant at the 5% level. As the bin corresponds to a 10% to 11% increase of the dividends from the previous year, it also looks like a convenient orientation value for possible future pay-out increases. The GRPV 11 Actually, there are seven cases in which dividends, repurchases and issuances sum up to exactly zero, but only pairwise - in none of them all three at the same time. test value in bin(10) of the variable D_NSCF_NSCF is 0.33, limiting previous reasoning to cash dividend pay-outs only. Focusing back on central bins, in Table 5 we investigate statistically significant (a 5% level is tested) differences between mean (median) values of discretionary accruals from the modified Jones model across bins. For each variable, with and without zeros, mean and median discretionary accruals from the model were computed for each bin in range from (-10) to (10), representing ±5% of lagged total assets or ±10% of lagged pay-outs, the difference due to different bin widths in the two approaches. Only the values for bins from (-2) to (2) are tabulated. We do this, firstly, because this is where our research interest lies as these are the most likely places in the distributions of pay-outs where the discretionary component of accruals would be important. Secondly, because there are not many significant differences further away from the centres of distributions. Finally, we keep our analyses compact for brevity of exposition. Bin means (medians) of discretionary accruals are compared to the means (medians) of discretionary accruals in the next bin using a t-test for the means and the Wilcoxson rank-sum test for the medians. For example, a boldfaced mean of DIV in bin(0) (0.955) indicates that it is significantly different from the mean in bin(1) (0.032). Similarly, a boldfaced median for NSCF in bin(-1) (0.011) indicates that it is significantly different from the median in the following bin(0) (0.091). Note that seemingly missing values are actually excluded for clarity. Variable DIV does not have negative bin observations (no negative dividends), while the results for bins (-2), (1) and (2) are not listed for versions of variables without zero observations because they are exactly the same as on the left-hand version. The versions only differ in the number of observations in the central bin (bin(0)) and possible differences only arise in comparisons of bin(-1) to bin(0) and of bin(0) to bin(1). In almost all instances significant differences in both means and medians of discretionary accruals are found at bin(0) or bin(-1) - the two that compare the central bin(0) with the neighbouring bins. Bin means of discretionary accruals are generally much larger than medians of discretionary accruals as a consequence of skewed distributions and are usually biggest in bin(0), means of bin(0) in first four variables being much bigger than means of other bins. Interestingly, excluding zero observations results in smaller bin(0) mean and median discretionary accruals compared to cases with all observations included and with the last two variables (D_DIV_DIV and D_NSCF_NSCF) they even become insignificantly different to other bins' means and medians. Assuming that discretionary accruals are associated with some form of purposeful managerial actions, and may be a tool to manage earnings or some other operating result by the management, their size and significance in central zero bins of distribution is at least indirect evidence of such actions. Table 5: Means and medians of discretionary accruals by bins Bin Mean Median Mean Median Mean Median Mean Median DIV DIV (without 0) NSCF NSCF (without 0) -2 -1 0.169* 0.114* 0.056* 0.011* 0.114* 0.011* 0 0.955* 0.128* 0.006* -0.025* 0.699* 0.091* 0.011 -0.026* 1 0.032 -0.012 0.027 -0.007 2 0.038 -0.008 0.028 -0.010 D_DIV D_DIV (without 0) D. _NSCF D_NSCF (without 0) -2 0.025* -0.030 0.063* -0.018 -1 0.003* -0.035* 0.003 -0.035* 0.028* -0.020* 0.028 -0.020 0 0.564* 0.034* 0.005* -0.023* 0.347* 0.004* 0.025 -0.022* 1 0.049* 0.006* 0.035* -0.007* 2 0.099 0.028* 0.071 0.007 D_DIV_ DIV D_DIV_DIV (wo 0) D_NSCF_NSCF D_NSCF_NSCF (wo 0) -2 -0.012 -0.039 0.056 -0.028 -1 -0.022* -0.051* -0.022 -0.051 0.099 -0.015* 0.099 -0.015 0 0.054* -0.016* -0.023 -0.043 0.102 0.007* -0.002 -0.036 1 -0.007 -0.037 0.051 -0.038 2 -0.007 -0.035 0.015 -0.033 Notes: This table reports means and medians of discretionary accruals form the modified Jones model by central bins of distributions. Each variable has bin means reported in the first column and bin medians in the second column of its box and is firstly evaluated with all observations included and then with zero observations excluded ("wo 0" in the last variable row stands for "without o"). Bolded font and asterisk denote that respective means (medians) are different from means (medians) in the following bin at the 5% significance level, i.e., a bolded* mean (median) in bin(0) is different from the mean (median) in bini at 5%. Tests used were t-test for means and Wilcoxson rank-sum test for medians. Modified Jones model is of the form: NDACC = afi/TAJ + a1(&REVt - &REC) + a3(PPE) . NDACC are non-discretionary accruals, TAt-l are lagged total assets (WC02999), AREVt is the change in revenues (WC01001), scaled by lagged total assets, ARECt is the change in receivables (WC02051), scaled by lagged total assets and PPEt is gross property plant and equipment (WC02301), scaled by lagged total assets. To estimate a1, a and a3 total accruals are considered as the dependent variable and calculated as: TACC = (ACAt - ACLt -ACasht + ASTDt - Dep)TAt-1. ACAt is the change in current assets (WC02201), ACLt is the change in current liabilities (WC03101), ACasht is the change in cash and cash equivalents (WC02001), ASTDt is the change in short term debt (WC03051), Dept are depreciation and amortization charges (WC01151) and TAt-1 are lagged total assets. Finally, discretionary accruals are obtained as the difference between total accruals and non-discretionary accruals predicted by the model. DIV = dividends (WC04551) scaled by lagged total assets; NSCF = (dividends + repurchases (WC04751) - issuances (WC04251)) = net shareholder cash flows scaled by lagged total assets; D_DIV = change in dividends from year (t-1) to (t) scaled by lagged total assets; D_NSCF = change in net shareholder cash flows from year (t-1) to (t) scaled by lagged total assets; D_DIV_DIV = change in dividends from year (t-1) to (t) scaled by lagged dividends; D_NSCF_NSCF = change in net shareholder cash flows from year (t-1) to (t) scaled by lagged net shareholder cash flows. For variables DIV, NSCF, D_DIV and D_NSCF bin width is 0.005 with lower bound inclusion, i.e., "zero bin" includes x if 0 < x < 0.005, "bin one" includes x if 0.005 < x < 0.01 etc., and for variables D_DIV_DIV and D_NSCF_NSCF bin width is 0.01 with lower bound inclusion, i.e., "zero bin" includes x if 0 < x < 0.01, "bin one" includes x if 0.01 < x < 0.02 etc. Bins in the range from -10 to 10 were tested but are not tabulated. Mean and median results of variables without zero observations are also not reported for bins -2, 1 and 2, as they are the same as in with zero observations included (the two versions differ only in the frequency of the zero bin). The two signals combined, that of accruals and breaks in pay-out distributions, indicate that the thresholds identified in this study can be associated with some firms' management activity. As firms aim to meet their planned, announced or established levels of pay-out on one side, and face anticipations of shareholders and potential investors on the other side, thresholds in form of positive pay-outs or pay-out changes gain in importance. Not wanting to fail expectations firms may make use of accrual manipulation to arrive at desired financial results that enable a suitable pay-out policy. 5. ADDITIONAL TESTS To address the potential sensitivity of discontinuity tests to neighbouring bin values suggested by previous research (Bennet & Bradbury, 2007), we first recalculate GRPV test statistics using two adjacent bins on either side of bin(0) (i.e., bins -2, -1, 1 and 2) and report it in column 1 of Table 6. The only difference to the main test is that the break in NSCF without zero observations is now only significant at 5% compared to previous 1% significance. All the other variables' rvalues are very similar to previously reported ones. We also re-calculate the GRPV test using only next-to-adjacent bins (i.e., -2 and 2) and the results (not tabulated) remain quantitatively and qualitatively substantially unchanged. This confirms the robustness of earlier our results to the details of test specifications. Extending the analysis beyond the primary hypotheses, we then use the test statistics to study what happens with the breaks in the distributions in relation to specified cutoffs, identified as potentially important for pay-out time dynamics. In columns 2 and 3 of Table 6 we look at the pre- and post- 2008 financial crisis periods. The inferences are unchanged with an adjustment in significance to 5% for NSCF and D_DIV_DIV, both without zero observations. What we do observe comparing the two sub-periods is that for the years 2008 and following all test values are smaller, mainly in the order of one half, than in pre-2008 period (apart from DIV and D_NSCF_NSCF, both without zero observations, which are insignificant as in the main test specification). Smaller values imply a less pronounced break in the distribution (although still highly significant) meaning less observations are concentrated in zero bins and more in the adjacent bins. This could be interpreted as some of the firms not pursuing or not being able to pursue pay-out thresholds in the crisis period, given the harsher economic conditions they found themselves in. Table 6: Additional GRPV discontinuity of distribution tests (1) (2) (3) (4) (5) (6) (7) 4 bins crisis effect ifrs used finance act used pre-2008 20088^ no yes pre-1997 1997&^ DIV 349.19 297.96 251.13 274.89 278.36 74.53 400.21 DIV (wo 0) 2.28 0.82 1.61 1.03 1.07 2.97 0.23 NSCF 124.50 90.59 75.17 86.61 78.64 33.14 112.96 NSCF (wo 0) 8.39 9.88 5.62 8.94 7.10 0.19 12.75 D_DIV 401.72 281.52 201.50 261.58 227.93 109.76 335.98 D_DIV (wo 0) 155.96 114.15 47.71 107.18 62.14 68.41 103.21 D_NSCF 158.15 108.84 61.19 105.74 65.54 51.50 115.40 D_NSCF (wo 0) 86.81 58.07 23.54 56.50 27.28 33.03 53.32 D_DIV_DIV 78.00 69.00 24.20 66.44 30.39 52.07 52.27 D_DIV_DIV (wo 0) 23.69 20.58 5.56 19.44 8.42 18.54 12.57 D_NSCF_NSCF 39.52 36.85 15.56 36.82 15.64 27.51 29.31 D_NSCF_NSCF (wo 0) 1.75 1.41 1.77 1.46 1.47 2.83 0.31 Notes: Reported are T values of GRPV discontinuity of distribution tests for zero bins of variables analysed with and without zero observations (the latter denoted by "wo 0" abbreviation). Column 1 reports statistics using 2 adjacent bins on either side of bin(0), columns 2 and 3 use 2008 as a cut-off year to analyse the effect of financial crisis, columns 4 and 5 analyse the effect of IFRS and column 6 and 7 use 1997 as a cut-off year to analyse the effect of change in legislation (Finance Act). With H00: the distribution function is continuous, the values of T at standard levels of significance are: at 10% |t| = 3.16, at 5% |t| = 4.47, and at 1% |t| = 10. As the number of observations in adjacent bins is required by the test, in the first two rows (case of DIV) bins on the left of zero (negative bins) are empty (there are no negative dividends) and are as such affecting test statistic computation. DIV = dividends (WC04551) scaled by lagged total assets (WC02999); NSCF = (dividends + repurchases (WC04751) - issuances (WC04251)) = net shareholder cash flows scaled by lagged total assets; D_DIV = change in dividends from year (t-1) to (t) scaled by lagged total assets; D_NSCF = change in net shareholder cash flows from year (t-1) to (t) scaled by lagged total assets; D_DIV_DIV = change in dividends from year (t-1) to (t) scaled by lagged dividends; D_NSCF_NSCF = change in net shareholder cash flows from year (t-1) to (t) scaled by lagged net shareholder cash flows. Our next cut-off is IFRS implementation. International Financial Reporting Standards and their predecessors, International Accounting Standards, are mainly regarded as being of higher quality than existing local standards (e.g., Barth, Landsman, & Lang, 2008; Armstrong et al., 2010), although alternative views are also not uncommon (Soderstrom & Sun, 2007; Ahmed, Neel, & Wang, 2013), and they also directly affected accounting for dividends as noted under sample selection. IFRS are compulsory since 2005 and this appears as a ready candidate for assessing the standards' effects. We deem it a second-best option as before 2005 firms could voluntarily adopt IFRS and even after 2005 data shows some financial statements in our sample as being prepared under UK GAAP. Our database allows us to identify the standards which the company used in preparing its reports and we thus classify 7,678 observations as prepared under IFRS. These mainly coincide with the period after 2005, but there is some overlapping with local standards, especially in years 2004-2007. The results (columns 4 and 5 in Table 6) in terms of subsample comparisons are analogous to that for the crisis effect. IFRS observations exhibit notably lower test values than non-IFRS observations for all but two insignificant variables lead- ing us to conjecture that IFRS usage is associated with "smoother" distributions. Potential explanation for this is the negative effect of stricter standards on firms' willingness and/or ability to achieve pay-out thresholds, positioning less of them in central bin(0). We identify the last cut-off to be 1997 as pointed out by the dividend taxation literature. Namely, in order to end the discriminatory tax treatment in favour of dividend pay-outs compared to capital gains, the Finance Act of 1997 increased taxation of dividend income, primarily affecting pension funds that were the largest class of investors in UK equities.12 Consequently, Bell & Jenkinson (2002) find a significant reduction in valuation of dividend income after the tax reform and initial evidence of reductions in dividend pay-out ratios, whereas Bond, Devereux & Kleem (2005) observe that it was the form of dividend payment that changed with the level marginally affected. Our two subsamples comparison in columns 6 and 7 of Table 6 reveals considerably smaller (yet again, still above critical values) values of discontinuity tests for most of the significant variables in the pre-1997 years compared to the later period. A potential explanation would be that after the 1997 tax reform dividends were not as large as before but still present (due to other investors' interests, signalling and other reasons), which resulted in their concentration in the smallest positive bin(s) of our distributions, producing a higher value of the test statistic. It has to be acknowledged however, that all these additional tests analyse only a specific factor possibly affecting pay-out dynamics and that firms' distributional decisions in real life are based on many elements, relative importance of which are changing in time. Moreover, even in our cases, there are overlapping effects especially towards the end of analysed time period. 6. conclusion This paper investigates the existence of pay-out-related thresholds as an extension of documented earnings management thresholds. With dividends and distinct net shareholder cash flows defined variables, discontinuities in their distributions are statistically analysed, employing a robust test that does not assume that the distributions of underlying variables are normal. The importance of pay-out policy for the firms' economic environment and for the firms' themselves (as a signalling mechanism, clientele and tax induced decisions etc.) leads us to consider threshold analysis to be of considerable importance for our study. We find evidence of breaks in distributions at suggested thresholds of zero or zero change of variables in question, supporting our reasoning that these are important for firms. Dividend thresholds are more pronounced than net shareholder cash flows thresholds suggesting the dominating role of cash dividends over share buybacks and over the netting role of new shares' issuances. Although repurchases are almost as common as dividend pay-outs, their effect is much smaller. Adding share issuances in the calculation to 12 More specifically, the 1997 Finance Act abolished repayment of dividend tax credits to tax-exempt investors, UK pension funds being the largest beneficiaries of the previous regulation. arrive at net shareholder cash flows disperses the pay-out distributions and reduces the breaks. Hence, repurchases and issuances are relatively much less important drivers of targeted pay-out level in the broader sense and net shareholder cash flows do not represent a separate threshold independent of cash dividends. Discretionary accruals as a proxy for earnings management are analysed at identified thresholds. We find significant differences and/or magnitudes of discretionary accruals at or in the closest proximity of central bins of distributions. This is another sign of their importance for firms as accruals are considered as a convenient and potentially strong earnings management tool. Additional analyses employ the discontinuity test to examine various sample partitions to arrive at more insightful results. We also find that a 10% dividend increase in the dividend paid is significant, suggesting the increase of dividends of 10% is common. Known caveats relate to distributional analysis being questioned as an earnings management measure and, although supportive of our hypotheses and considered general, the accrual model employed is merely one of several accruals modes and these have been found to produce results of different significance or even conclusions. In a related, but not directly comparable research, Dechow, Richardson & Tuna (2003) are not able to confirm that discretionary accruals are driving the breaks in earnings distributions and offer supplementary explanations. Nevertheless, we consider the evidence in this paper strong enough to stress the importance of firms' pay-out policy, shedding additional light on the effects of pay-out policy components. Finally, we also identify some potential areas for future research. For example, it might be possible to derive more precise tests that would be able to distinguish the effects of the financial crisis and the effects of new standards, where the two periods overlap significantly. This might be related to the use of more refined discretionary accruals models. These models might also be investigated independently of the breaks due to standards, financial crisis, etc. We also do not consider possible "real" earnings management (Roy-chowdhury, 2006), which might be a significant component of the overall management to achieve earnings and net shareholders' flows thresholds. references Ahmed, A. S., Neel, M., & Wang, D. (2013). Does mandatory adoption of IFRS improve accounting quality? Preliminary evidence. Contemporary Accounting Research, forthcoming. Armstrong, C. S., Barth, M. E., Jagolinzer, A. D., & Riedl, E. J. (2010). Market reaction to the adoption of IFRS in Europe. The Accounting Review, 85 (1), 31-61. Arya, A., Glover, J., & Sunder, S. (1998). Earnings management and the revelation principle. Review of Accounting Studies (3), 7-34. Atieh, A., & Hussain, S. (2012). Do UK firms manage earnings to meet dividend thresholds? Accounting and Business Research, 42 (1), 77-94. Baker, H. K., Powell, G. E., & Veit, E. T. (2002). Revisiting the dividend puzzle: Do all of the pieces now fit? Review of Financial Economics, 11, 241-261. Baker, M., & Wurgler, J. (2004). A catering theory of dividends. The Journal of Finance, 59 (3), 1125-1165. Barth, M. E., Landsman, W. R., & Lang, M. H. (2008). International accounting standards and accounting quality. Journal of Accounting Research, 46 (3), 467-498. Beaver, W. H., McNichols, M. F., & Nelson, K. K. (2007). An alternative interpretation of the discontinuity in earnings distributions. Review of Accounting Studies, 12, 525-556. Bell, L., & Jenkinson, T. (2002). New evidence of the impact of dividend taxation and on the identity of the marginal investor. The Journal of Finance, 57 (3), 1321-1346. Bennet, B., & Bradbury, M. E. (2007). Earnings thresholds related to dividend cover. Journal of International Accounting Research, 6 (1), 1-17. Bergstresser, D., & Philippon, T. (2006). CEO incentives and earnings management. Journal of Financial Economics, 80, 511-529. Bhattacharyya, N. (2007). Dividend policy: A review. Managerial Finance, 33 (1), 4-13. Bond, S. R., Devereux, M. P., & Kleem, A. (2005). Dissecting dividend decisions: Some clues about the effect of dividend taxation from recent UK reforms. IFS Working Papers, 05/17. London, UK: Institute for Fiscal Studies. Brav, A., Graham, J. R., Harvey, C. R., & Michaely, R. (2005). Payout policy in the 21st century. Journal of Financial Economics, 77, 483-527. Brigham, E. E., & Ehrhardt, M. C. (2005). Financial management: Theory and practice (11th ed.). Mason: Thomson South-Western. Burgstahler, D., & Dichev, I. (1997). Earnings management to avoid earnings decreases and losses. Journal of Accounting and Economics, 24, 99-126. Carslaw, C. A. (1988). Anomalies in income numbers: Evidence of goal oriented behaviour. The Accounting Review, 63 (2), 321-327. Daniel, N. D., Denis, D. J., & Naveen, L. (2008). Do firms manage earnings to meet dividend thresholds? Journal of Accounting and Economics, 45, 2-26. DeAngelo, H., & DeAngelo, L. (2006). The irrelevance of the MM dividend irrelevance theorem. Journal of Financial Economics, 79, 293-315. Dechow, P. M., Richardson, S. A., & Tuna, I. (2003). Why are earnings kinky? An examination of the earnings management explanation. Review of Accounting Studies, 8, 355-384. Dechow, P. M., Sloan, R. G., & Sweeney, A. P. (1995). Detecting earnings management. The Accounting Review, 70 (2), 193-225. Dechow, P., Ge, W., & Schrand, C. (2010). Understanding earnings quality: A review of the proxies, their determinants and their consequences. Journal of Accounting and Economics, 50, 344-401. Degeorge, F., Patel, J., & Zeckhauser, R. (1999). Earnings management to exceed thresholds. Journal of Business, 72 (1), 1-33. Denis, D. J., & Osobov, I. (2008). Why do firms pay dividends? International evidence on the determinants of dividend policy. Journal of Financial Economics, 89, 62-82. Durtschi, C., & Easton, P. (2005). Earnings management? The shapes of the frequency distributions of earnings metrics are not evidence ipso facto. Journal of Accounting Research, 43 (4), 557-592. Durtschi, C., & Easton, P. (2009). Earnings management? Erroneous inferences based on earnings frequency distributions. Journal of Accounting Research, 47 (5), 1249-1281. Erickson, M., & Wang, S.-w. (1999). Earnings management by acquiring firms in stock for stock mergers. Journal of Accounting and Economics, 27 (2), 149-176. Fama, E. F., & French, K. R. (2001). Disappearing dividends: Changing firm characteristics or lower propensity to pay? Journal of Financial Economics, 60, 3-43. Garrod, N., Ratej Pirkovic, S., & Valentincic, A. (2006). Testing for discontinuity or type of distribution. Mathematics and Computers in Simulation, 71, 9-15. Grullon, G., & Michaely, R. (2002). Dividends, share repurchases and the substitution hypothesis. The Journal of Finance, 57 (4), 1649-1684. Grullon, G., Michaely, R., & Swaminathan, B. (2002). Are dividend changes a sign of firm maturity? Journal of Business, 75 (3), 387-424. Handley, J. C. (2008). Dividend policy: Reconciling DD with MM. Journal of Financial Economics, 87, 528-531. Hayn, C. (1995). The information content of loses. Journal of Accounting and Economics, 20, 125-153. Hribar, P., Jenkins, N. T., & Johnson, W. B. (2006). Stock repurchases as an earnings management device. Journal of Accounting and Economics, 41, 3-27. International Accounting Standards Board. (2010). IAS 10 - Events after the reporting period. Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica, 47 (2), 263-291. Kasanen, E., Kinnunen, J., & Niksanen, J. (1996). Dividend-based earnings management: Empirical evidence from Finland. Journal of Accounting and Economics, 22, 283-312. Lintner, J. (1956). Distributions of income of corporations among dividends, retained earnings and taxes. The American Economic Review, 46 (2), 97-113. Miller, R., & Modigliani, F. (1961). Dividend policy, growth and the valuation of shares. Journal of Business, 34, 411-433. Moir, L., & Sudarsanam, S. (2007). Determinants of financial covenants and pricing of debt in private debt contracts: The UK evidence. Accounting and Business Research, 37 (2), 151-166. Ohlson, J. A. (1995). Earnings, book values and dividends in equity valuation. Contemporary Accounting Research, 11 (2), 661-687. Renneboog, L., & Trojanowski, G. (2005). Patterns in payout policy and payout channel choice of UK firms in the 1990s. Discussion paper, 2005-22, 1-55. Tilburg University, Center for Economic Research. Ronen, J., & Yaari, V. (2010). Earnings management: emerging insights in theory, practice, and research. New York: Springer. Roychowdhury, S. (2006). Earnings management through real activities manipulation. Journal of Accounting and Economics, 42, 335-370. Scott, D. W. (1979). On optimal and data-based histograms. Biometrika, 66 (3), 605-610. Shefrin, H. M., & Statman, M. (1984). Explaining investor preference for cash dividends. Journal of Financial Economics, 13, 253-282. Skinner, D. J. (2008). The evolving relation between earnings, dividends and stock repurchases. Journal of Financial Economics, 87, 582-609. Soderstrom, N. S., & Sun, K. J. (2007). IFRS adoption and accounting quality: A review. European Accounting Review, 16 (4), 675-702. The Institute of Chartered Accountants in England and Wales. (1980). SSAP 17 - Accounting for post balance sheet events. Tong, Y. H., & Miao, B. (2011). Are dividends associated with the quality of earnings? Accounting Horizons, 25 (1), 183-205. Wand, M. P. (1997). Data-based choice of histogram bin width. The American Statistician, 51 (1), 59-64.