The Ability of Fundamental Analysis to predict
Future Earnings and to earn Abnormal Returns: An
Empirical Investigation from Western European
Markets
Master’s Thesis MSc IFM
Erhard Dubs
University of Groningen: S2557177 Uppsala University: 890504-‐P532
Supervisor: Dr. Ing. N. Brunia Assessor: Dr. H. Vrolijk Date: 29.01.2015
University of Groningen Faculty of Economics and Business MSc International Financial Management
Uppsala University
Department of Business Studies MSc Business and Economics
II
ABSTRACT
This paper investigates whether fundamental signals can predict future earnings changes and whether capital markets incorporate these predictions properly into current stock price. Fundamental signals used in this research are selected based on a theoretically guided approach. My sample consists of 6930 firm year observations listed in Western Europe between 2000 and 2011. The findings support my first hypothesis that fundamental signals are useful in predicting future earnings, although the effect of some signals is not in line with expectations as some of them are related to the opposite direction. My findings are not entirely consistent with previous literature. Additional tests, where predictive power of fundamental signals are measured on an aggregate level, underscore the validity of each fundamental signal used in this research. The relationship between fundamental signals and future abnormal returns provide no support for my second hypothesis, as most signals are statistically insignificant. Even though the fundamental trading strategy yield to 7.2% average abnormal returns, the observed abnormal returns are statistically insignificant. Various tests demonstrate the robustness of the findings.
Keywords: Fundamental analysis, earnings prediction, abnormal returns, fundamental
trading strategy, market efficiency, contextual analysis.
Data Availability: all data are available from public sources.
TABLE OF CONTENTS
1. Introduction ... 1
2. Literature ... 3
2.1 Literature on fundamental analysis ... 3
2.2 Fundamental signals and control variables ... 6
2.3 Hypotheses ... 9
3. data and Methodology ... 10
3.1 Sample ... 10
3.2 Data ... 12
3.3 Method ... 15
4. Results ... 18
4.1 Relation between Fundamental Signals and Future Earnings ... 18
4.2 Relation between Fundamental Signals and Abnormal Returns ... 22
5. Robustness checks ... 26
5.1 Current firm performance context ... 26
5.2 Industry context ... 27
5.3 Country context ... 28
5.4 Control of outlier treatment ... 28
5.5 Control of nonlinear relations ... 29
6. Conclusion ... 30
References ... 32
Appendices ... 35
IV
LIST OF TABLES
Table 1: Summary of hypothesized relations and findings of Abarbanell and Bushee ... 8
Table 2: Descriptive Statistic ... 11
Table 3: Country and Industry Breakdown of Observations ... 12
Table 4: Definition and Description of Raw Data ... 13
Table 5: Definitions and Measurements of Variables ... 14
Table 6: Regression of One-‐Year-‐Ahead Changes in Earnings ... 20
Table 7: Regression of One-‐Year-‐Ahead Changes in Earnings on Composite Score ... 22
Table 8: Regression of Abnormal Returns ... 24
Table 9: Annual Abnormal Returns ... 26
Table 10: Summary of results ... 31
APPENDICES Appendix A: Regression of Long-‐Term Abnormal Returns to Fundamental Signals ... 35
Appendix B: Regression of One-‐Year-‐Ahead Changes in Earnings conditioned for Firm Performance ... 36
Appendix C: Regression of One-‐Year-‐Ahead Changes in Earnings conditioned on Industry Membership ... 37
Appendix D: Regression of One-‐Year-‐Ahead Changes in Earnings conditioned on Country Membership ... 38
Appendix E: Regression of One-‐Year-‐Ahead Changes in Earnings without adjustments for Outlier ... 39
Appendix F: Regression of One-‐Year-‐Ahead Changes in Earnings to test for Nonlinear Relations ... 40
1. INTRODUCTION
Accounting based earning prediction and stock anomalies have a long history in literature. The main motivation for fundamental analysis is to identify mispriced securities by considering financial statements and other factors affecting a firm’s value (Kothari 2001). An important research line is multivariate fundamental analysis, which has its origin in the research of Ou and Penman (1989a). They combine a large set of accounting signals to predict the sign on the changes of future earnings. Building on this idea, Lev and Thiagarajan (1993) use fundamental signals that refer to a specific configuration of several fundamental variables, which are popular amongst analysts and can be intuitively motivated, instead of relying on an extensive list of signals. Follow-‐up research, such as Abarbanell and Bushee (1997, 1998), demonstrates that the information contained in Lev and Thiagarajan’s (1993) set of fundamental signals is not fully utilized by the stock market and hence an investment strategy based on these signals may yield abnormal returns. However, neither Lev and Thiagarajan (1993) nor Abarbanell and Bushee (1997, 1998) pretend to offer an optimal set of fundamental signals. Therefore, alternative and perhaps complementary signals could lead to a more accurate prediction of future earnings and could increase abnormal returns (Seetharam and Auret 2014).
Fundamental analysis is relevant for scholars, practitioners as well as regulators. The chief motivation for scholars is to test market efficiency by assessing how financial information influences security prices. Practitioners, such as portfolio manager, are interested in identifying security mispricing to earn abnormal returns. Regulators are focused on whether capital markets efficiently value financial reporting standards. An example would be: “does the IFRS accounting convey new information to market participants?” (Kothari 2001).
2
provide incremental explanatory power to explain future earnings changes over current earnings changes. In the second step, I investigate the relationship between fundamental signals and future abnormal returns by forming zero-‐investment portfolios that exploit information in the fundamental signals. The relationship between fundamental signals and future earnings changes after conditioning for current firm performance, industry and country membership is also examined. The sample of this research is based on 6930 firm year observations in Western Europe between 2000 and 2011.
This research enriches the literature in several ways. First of all, whereas previous research is restricted to North American data, this paper investigates Western European data. Similar to the US, Western European capital markets are highly developed, but are grounded on different accounting standards and corporate government systems. Europe is characterized by the IFRS financial disclosure, that is why European financial statement information might be utilized differently by market participants. In addition, differences between corporate governance systems could vary in how firms account for their economic activities in the US and Western Europe. I expect that such variations have also an impact on single Western European countries. Second, most of research has been conducted using overlapping time periods in the 1980s and 1990s. This research investigates the 2000s that are characterized by dynamic movements in the capital markets. According to Swanson, Ress and Juarez-‐ Valdes (2003), investors’ need for forward-‐looking accounting information is greater during turbulent economic times. Hence, this research examines the robustness of earlier studies. Another important contribution to literature is the extension of the fundamental analysis model suggested in Lev and Thiagarajan (1993) with three additional earnings-‐relevant fundamental signals motivated by economic reasoning.
returns provide no support for my second hypothesis, as most signals are statistically insignificant. Moreover, the overall average abnormal return of 7.2% of the fundamental trading strategy is insignificant.
The remainder of the paper is structured as follows: section 2 discusses previous literature concerning fundamental analysis, describes the fundamental signals and establishes the hypotheses. Section 3 describes the sample, source of the data and research method. Key findings are reported and discussed in section 4. Section 5 provides additional robustness checks and section 6 presents the conclusion.
2. LITERATURE
2.1 Literature on fundamental analysis
4
18 signals from a large set of signals by data reduction, is a highly controversial way of research.
Lev and Thiagarajan (1993) used a more theoretically guided approach to select fundamental signals by studying which signals are repeatedly referred to in financial statement analysis texts and analysts’ reports. These leads to 12 fundamental signals, where a signal refers to a specific configuration of several fundamental variables. These fundamental signals measure the unexpected changes in inventory, accounts receivable, capital expenditure, gross margin, selling and administrative expenses, provision for doubtful receivables, effective tax rate, order backlog, labor force productivity, earnings quality and in an indicator variable which identifies an auditor qualification. According to Lev and Thiagarajan (1993), these signals provide an indicator of persistence (also referred to as “quality”) and growth of earnings. Unlike Ou and Penman (1989a), their research aims to explain the value relevance of fundamental signals by using stock price as a benchmark. Applying contemporaneous stock returns as a dependent variable assumes that markets fully reflect financial statement information and thus implies that market prices are efficient. Their findings show that most of the fundamental signals are statistically significant in their hypothesized direction to explain contemporaneous stock returns. Thus indicate that investors take the predicting nature of signals into account. This research had a notable impact on following fundamental research by establishing a set of intuitively motivated and practically grounded fundamental signals. However, the analysis of Lev and Thiagarajan (1993) relies on the not verified assumption that contemporaneous stock returns capture the full extent of financial statement information.
revisions are the difference between the forecasts after and before the announcement of earnings. Their results show that analysts forecast revisions fail to reflect all the information about future earnings contained in these signals, although many signals have predictive power to explain analysts forecast revisions. This findings provide the possibility that some of the predicting nature of fundamental signals has been unutilized by the market.
In their follow-‐up paper, Abardanell and Bushee (1998) utilize their prior findings as an investment strategy to earn abnormal returns. The idea is based on the fact that if analysts underreact to information in signals, there is a chance that markets in general are also unable to utilize this information. The approach of Abardanell and Bushee (1998) is different to Ou and Penman (1989a) by maintaining a focus on individual fundamental signals, instead of combining them into a single summary measure. The method of Abardanell and Bushee (1998) has the advantage that they can evaluate the robustness of each fundamental signals’ predictive power. Their findings show a statistically significant average 12-‐month cumulative size-‐adjusted abnormal return of 13.2% over the sample period 1974-‐1988. Furthermore, they test the effect of fundamental signals on longer term abnormal returns. This test shows that after the 12-‐ month period no significant abnormal returns are earned, and hence the diminishing abnormal returns serve as evidence that observed abnormal returns are not simply a premium for some risk.
6
composite score, which was initially established by Lev and Thiagarajan (1993), where for each positive (negative) signal value a 1 (0) is assigned. A high (low) score denote for a fundamentally strong (weak) firm. However, their approach also summarizes the signals into one measure and hence cannot prove the validity of each signal. It is worth mentioning that the set of fundamental signals used in Piotroski (2000) is based on solid economic reasoning, but almost totally different from the one used in Lev and Thiagarajan (1993). This points out that there is still space for signals that can be relevant to predict future earnings.
All in all, most of the research in this field is based on US data, although some exceptions exist. Particularly noteworthy are Al Debie and Walker (1999), who replicate Lev and Thiagarajan (1993) with some modifications to a UK-‐based perspective. Their findings largely confirm those of Lev and Thiagarajan (1993). Swanson et al. (2003) study in particular effects of inflation in Mexico by following the methodology of Abardanell and Bushee (1997, 1998) and indicate the effectiveness of fundamental signals. However, the samples used in these researches are somewhat small. Additionally, the different models indicate that there is a broad spectrum of variations that can be studied.
2.2 Fundamental signals and control variables
guidance to use an industry benchmark for capital expenditure. In contrast, using capital expenditure relative to depreciation to measure for the impact of investment activities is common practice, as it is an indicator of how much a firm is investing in its business. I expect that these signals can predict future changes in earnings and future abnormal returns.
1. An increase in inventory relative to sales is normally considered as a bad signal for future earnings by analyst. Such an occurrence may indicate difficulties in generating cash or that portion of inventory will become obsolete (Lev and Thiagarajan 1993). Although in some cases an growth in inventory might be interpreted as a good signal, since it could signal an increase in future sales and reduces the risk of inventory shortages, Lev and Thiagarajan (1993) findings confirm their assumption that a disproportional inventory increase is on average a bad signal.
2. Disproportionate increase in accounts receivable relative to sales is viewed as unfavorable signal, due to the fact that such an circumstance could indicate extensions of credit terms to maintain sales levels. A disproportionate accounts receivable increase might also imply earnings management, where unrealized revenues have been booked as sales, indicating lower earnings persistence (Lev and Thiagarajan 1993).
3. A disproportionate increase in the gross margin relative to sales is interpreted as a good sign for earnings, since it effects firm’s long-‐term performance and is therefore informative regarding earnings persistence. Underlying factors driving this relation are intensity of competition and the relation between variable and fixed costs (Lev and Thiagarajan 1993).
4. An increase in administrative (S&A) expenses in comparison with sales is viewed by analysts as a loss of cost control that cannot be passed on to customers, indicating lower future earnings. In addition, such an occurrence can be interpreted as an increased effort to produce sales that was not completely effective (Lev and Thiagarajan 1993).
8
6. Analysts generally appreciate restructuring, in particular labor force reductions. Although these reductions often lead to higher cost in the year of restructuring, in subsequent years such actions should increase earnings. Measuring sales by the number of employees allows to capture the changes in the efficiency of labor and the changes in the number of employees (Lev and Thiagarajan 1993).
7. Disproportionate increase in capital expenditure relative to depreciation is generally interpreted as a good sign for future earnings, since net investments provide benefits in generating future earnings. However, there are good reasons to believe that a cut in capital expenditure boost earnings in the short-‐run, due to the fact that capital projects do not immediately impact.
8. An increase in earnings before interest, taxes depreciation and amortization (EBITDA) in relation to net debt is considered as positive signal by analysts. Such an incidence may show facilitations in generating sufficient internal funds and hence ensures future financial flexibility (Myers and Majluf 1984). However, other economic theories, which favor external fund raising as a positive sign, could be reasonable (see e.g. Dimitrov and Jain 2008).
9. Disproportionate increase in current asset relative to current liabilities is considered as a good signal by analysts, since improved liquidity reflects firm’s ability to service debt obligations, and therefore is a measure of firm’s financial health (Piotroski 2000). According to Sloan (1996), financial health measurements are indicator of future earnings persistence. On the other hand, Abarbanell and Bushee (1997) argue that short-‐term orientated expansionary policies can promote short-‐term sales and earnings growth.
Table 1: Summary of hypothesized relations and findings of Abarbanell and Bushee
Independent Variables
Dependent Variable INV AR GM S&A ETR LFP NI LEV LIQ
+(-) + + + + + +(-) +(-) +(-) Abarbanell and Bushee’s findings EPSt+1 +** -** +** + +** +** Rt+1 +** + +** +** - +
Headers indicate hypothesized directions (alternative directions in parentheses) Table summarize observed relations
Statistical significance level are based on one tailed tests, where * marks a level of 0.10, ** marks a level of 0.05 and *** marks a level of 0.01
are included to distinguish whether returns to signals are based on risk-‐based explanations or earnings surprise reasons. This has also been the method to control for risk-‐based explanation in related literature (e.g. Abarbanell and Bushee 1997, Swanson et al. 2003). While Abarbanell and Bushee (1998) include equity beta in their regression to measure whether beta can explain partially observed abnormal returns, Swanson et al. (2003) add market capitalization and book-‐to-‐market ratio to their model. Finance literature says that each of these risk factors has explanatory power on excess stock returns. I include these three risk proxies in my regression to control for risk-‐based explanations of possible abnormal returns. Beta measures the risk of the equity of a firm that arise from market movements. A higher beta indicates stronger upward and downward movements of firm’s equity. Thus, a higher beta should lead to a higher required return. The book-‐to-‐market ratio signals firm’s relative distress. This means that more distressed firms with persistently low earnings tend to have a higher ratio, implying lower expectations on the future generation of cash flows, and hence investors require higher returns. Market capitalization also shows the riskiness of a firm, as firms with smaller market capitalization tend to generate larger risk-‐adjusted returns than larger firms (Banz 1981). This indicates that smaller firms are related to higher business risk. Following Abarbanell and Bushee (1998), the change in current earnings is also included as a control variable to measure whether the fundamental signals have added incremental explanatory power over current earnings changes.
2.3 Hypotheses
One goal of my research is to test the ability of my fundamental signals to predict future earnings among Western European public firms using an extended model. Therefore, I examine whether there is any relation between changes in one-‐year-‐ahead earnings and fundamental signals. The first hypothesis is:
H1: Fundamental signals predict future earnings in Western Europe.
Secondly, this research investigates whether a fundamental trading strategy based on these earnings predicting signals leads to abnormal return. Hence, the second hypothesis is:
H2: A fundamental trading strategy based on fundamental signals earns significant abnormal returns.
10
3. DATA AND METHODOLOGY 3.1 Sample
The sample is obtained from the Datastream database using companies identified in Bureau van Dijk's Orbis database after initial filtering for region, industry and quoted companies. I only focus on Western Europe (27 countries1) and exclude financial
institutions and the service sector (NACE Rev. 2 code 64 and upwards). The reason to exclude these sectors is due to their different accounting characteristic making the fundamental signals not applicable. Furthermore, only firms that use the calendar year as their reporting period are included in the sample, limiting the sample to 3,851 companies. I include delisted firms to avoid the possibility of survivorship biases. The accounting data period is from 2000 to 2011, allowing to conduct subsequent abnormal returns from 2002 to 2013. Sample firm with missing data will be excluded for that sample year. The MSCI Europe Index is used as the market index to calculate the abnormal returns, as this index covers approximately 85% of the free float-‐adjusted market capitalization across the European Developed Markets (MSCI 2014). Data is gathered in euro; however, the currency is irrelevant as variables are expressed as changes in percentages or ratios. All countries used in this research introduced IFRS financial disclosure for publicly traded firms in 20052.
The annual observations range from 307 (in 2000) to 909 (in 2011) firms per year, leading to a total sample size of 6930 firm year observations. The manufacturing industry is by far the largest sector in the sample with a share of 64%. The dominating countries in the sample are Germany and Great Britain, each covering 21% of the total sample. Table 2 shows the descriptive statistics for the observations between 2000-‐ 2011 and table 3 summarizes the industry and country composition of the observations.
Worth mentioning is the fact that on average security returns are higher than the market return in my sample, whereas average beta is smaller than 1. This indicates that the sample is unbalanced, which means that the sample is not a full representative of the market. The statistical distribution of fundamental signals also shows that my sample contains extreme values, which may bias the relation between dependent and
1 Western Europe countries include: Andorra, Austria, Belgium, Cyprus, Denmark, Finland, France,
Germany, Gibraltar, Greece, Iceland, Ireland, Italy, Lichtenstein, Luxemburg, Malta, Monaco, Netherlands, Norway, Portugal, San Marino, Spain, Sweden, Switzerland, Turkey, United Kingdom and Vatican.
2 Due to the fact that a lot of European firms introduced IFRS before it becomes mandatory, effects of IFRS
independent variable. Due to the fact that I am interested in the relation between fundamental signals and future earnings in regular conditions, I have truncated the sample at the 1% and 99% tails of each variable. Such an approach of dealing with outliers is a typical method in related literature (see, e. g. Lev and Thiagarajan 1993). This may allow drawing more meaningful conclusions in my first investigation. To avoid bias to test my second hypothesis, variables are not truncated in the second section. Outlier treatment is not crucial for the test of the trading strategy, due to fact that signals are ranked into deciles and weighted equally. As a result inclusion of outliers will not have significant impact on the results of the fundamental trading strategy.
Table 2: Descriptive Statistic
Variable Mean Median Std. Dev. Minimum Maximum Percentiles
25% 75% △EPSt+1 0.018 0.007 0.519 -37.732 5.847 -0.027 0.045 Rt+1 0.033 -0.015 0.464 -1.216 8.608 -0.231 0.209 INV -0.079 0.003 2.583 -184.018 4.882 -0.041 0.045 AR -0.001 0.005 0.235 -9.035 6.596 -0.026 0.034 GM -0.001 0.001 0.451 -16.358 6.128 -0.021 0.023 SG&A -0.002 0.001 0.196 -3.909 7.192 -0.022 0.025 ETR -0.017 -0.004 33.661 -2544.942 711.669 -0.074 0.030 LFP 0.200 0.033 8.621 -1.000 710.265 -0.043 0.117 NI 0.126 -0.005 3.014 -6.774 186.213 -0.085 0.083 LEV 0.003 0.007 11.650 -573.920 647.825 -0.156 0.158 LIQ -0.012 -0.002 0.164 -5.286 3.358 -0.040 0.032 △EPSt 0.029 0.008 0.254 -3.024 5.847 -0.025 0.047 BETA 0.771 0.710 0.665 -7.730 4.640 0.340 1.130 MCAP 6.248 6.162 2.263 -0.616 12.425 4.596 7.834 BM 0.779 0.610 1.022 -33.333 20.000 0.365 1.000
The sample is untruncated and consists of 6930 firm year observations between 2000 and 2011.
△EPSt+1: One-Year-Ahead Change in Earnings; Rt+1: Buy-and-Hold Abnormal Returns; INV: Inventory;
AR: Accounts Receivable; GM: Gross Margin; S&A: Selling, General and Administrative Expenses; ETR: Effective Tax Rate; LFP: Labor Force Productivity; NI: Net Investment; LEV: Leverage; LIQ: Liquidity; △EPSt: Current Change in Earnings; BETA: Beta; MCAP: Market Capitalization;
BM: Book-to-Market.
12
Table 3: Country and Industry Breakdown of Observations
Country NACE classification CH DE FI FR IT NL SE UK Othersa Total As % Agriculture (01-03) 0 2 0 0 13 0 0 61 6 82 1% Mining (05-09) 4 5 0 25 12 11 17 158 78 310 4% Manufacturing (10-33) 543 969 184 174 541 193 340 837 655 4436 64% Energy supply (35) 20 67 0 0 66 0 0 16 45 214 3% Water supply (36-39) 0 9 12 9 16 0 0 8 10 64 1% Construction (41-43) 1 19 2 2 51 5 46 124 52 302 4% Wholesale and retail trade (45-47) 44 108 25 18 15 41 42 103 59 455 7% Transportation (49-53) 24 62 15 1 63 9 0 28 98 300 4% Accommodation and food service (55-56) 1 1 0 0 12 0 4 33 19 70 1% Telecommunication (58-63) 45 196 11 34 152 34 42 99 84 697 10% Total 682 1438 249 263 941 293 491 1467 1106 6930 100% As % 10% 21% 4% 4% 14% 4% 7% 21% 16% 100%
a Other countries include :Andorra, Austria, Belgium, Cyprus, Denmark, Gibraltar, Greece, Iceland, Ireland,
Lichtenstein, Luxemburg, Malta, Monaco, Norway, Portugal, San Marino, Spain, Switzerland, Turkey, and Vatican
3.2 Data
The majority of fundamental signals used in this research is based on the signals (discussed in section 2.2) suggested by Lev and Thiagarajan (1993), but is calculated as suggested in Abarbanell and Bushee (1998). This simplifies the comparison between my results and those of Abarbanell and Bushee (1998), which I use because of the same methodology as a benchmark. This approach allows to expect a positive effect of fundamental signals on future earnings and abnormal returns, and makes later sign assimilations of fundamental signals unnecessary. Because of data availability issues, this research applies only 6 out of 12 fundamental signals created by Lev and Thiagarajan (1993). In addition, this research initiates three new fundamental signals, which are motivated by economic reasoning. This set aims to improve the model of Lev and Thiagarajan (1993), leading to a more accurate prediction of future earnings and could increase abnormal returns. Table 4 presents the definition and description of the
Table 4: Definition and Description of Raw Data
Symbol Name Description Code in
Datastream S Sales Gross sales and other operating revenue less discounts,
returns and allowances.
WC 01001 M Gross Margin The difference between sales or revenues and cost of goods
sold.
WC 01100 SG&A Selling, General &
Administrative Expenses
Expenses not directly attributable to the production process but relating to selling, general and administrative functions.
WC 01101 D Depreciation The allocation of cost of a depreciable asset to the accounting
periods covered during its expected useful life to a business excluding amortization and impairments on acquired intangibles.
WC 01148
EBT Pretax Income All income/loss before any federal, state or local taxes. Extraordinary items reported net of taxes are excluded.
WC 01401
TAX Income Taxes Income tax levied on the income of a company . WC 01451
EBITDA EBITDA The earnings of a company before interest expense, income taxes and depreciation and amortization.
WC 18198
C Cash The sum of cash and short-term investments. WC 02001
R Receivables The amounts due to the company resulting from the sale of goods and services on credit to customers.
WC 02051 FG Finished Goods The inventory of goods, which are ready for sale. WC 02099 I Inventories Tangible items acquired for either resale directly or included
in the production.
WC 02101 CA Current Asset Cash, inventories, accounts receivable and other assets sold
or consumed within one year or one operating cycle.
WC 02201 STD Short-term Debt Debt payable within one year including current portion of
long-term debt.
WC 03051 CL Current Liabilities Debt or other obligations that the company expects to satisfy
within one year.
WC 03101 LTD Long-term Debt All interest bearing financial obligations, excluding amounts
due within one year.
WC 03251 CAPEX Capital
Expenditures
The funds used to acquire fixed assets other than those associated with acquisitions.
WC 04601 P Share Price The closing price of the company's stock at the fiscal year
end.
WC 05001 EPS Earnings per Share The earnings of the company excluding extraordinary items
reported after tax from earnings per share.
WC 05201 EMP Employees The number of both full and part time employees of the
company.
WC 07011
M/B Market
Capitalization/ Common Equity
The total market capitalization of the company based on year end price and number of shares outstanding divided by the common equity of the company.
WC 09704
TR Total Return The growth of a share over a specified period, assuming that dividends are reinvested to purchase additional units of an equity at the closing price applicable on the ex-dividend date.
RI
MV Market Value The total market capitalization of the equity of the company based on price and number of shares outstanding on a specific date.
MV
14
In similar vein with Lev and Thiagarajan (1993), I use expected values for all fundamental signals, where these are recommended by Lev and Thiagarajan (1993). This enables to see the unexpected components of fundamental signals. The expected values are defined by Lev and Thiagarajan (1993) as 𝐸(𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒) = ½(𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒!–! +
𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒!–!). This expectation model is common practice in related literature. Table 5
shows the dependent variables, fundamental signals and control variables. Table 5: Definitions and Measurements of Variables
Symbol Name Measurement
Dependent Variables △ EPS!!! One-Year-Ahead Change in Earnings 𝐸𝑃𝑆!!!– 𝐸𝑃𝑆! 𝑃!–! 𝑅!!! Buy-and-Hold Abnormal Returns 𝑇𝑅!,!!!– 𝑇𝑅!,!!! Fundamental Signals INV Inventory 𝑆!– 𝐸(𝑆) 𝐸(𝑆) – 𝐹𝐺!– 𝐸(𝐹𝐺) 𝐸(𝐹𝐺) AR Accounts Receivable 𝑆!– 𝐸 𝑆 𝐸 𝑆 – 𝑅!– 𝐸 𝑅 𝐸 𝑅 GM Gross Margin 𝑀!– 𝐸(𝑀) 𝐸(𝑀) – 𝑆!– 𝐸(𝑆) 𝐸(𝑆)
S&A Selling, General and Administrative Expenses 𝑆!– 𝐸(𝑆) 𝐸(𝑆) – 𝑆𝐺&𝐴!– 𝐸(𝑆𝐺&𝐴) 𝐸(𝑆𝐺&𝐴)
ETR Effective Tax Rate [𝑇𝑅!– (!
! 𝑇𝑅!–!)]𝑥 ! !!! △ 𝐸𝑃𝑆!; where 𝑇𝑅!=!"#!"#! ! LFP Labor Force Productivity 𝑆! 𝐸𝑀𝑃!– 𝑆!!! 𝐸𝑀𝑃!!–!)/ 𝑆!–! 𝐸𝑀𝑃!!–! NI Net Investment 𝐶𝐴𝑃𝐸𝑋!– 𝐸 𝐶𝐴𝑃𝐸𝑋 𝐸 𝐶𝐴𝑃𝐸𝑋 – 𝐷!– 𝐸 𝐷 𝐸 𝐷 LEV Leverage 𝐸𝐵𝐼𝑇𝐷𝐴!– 𝐸(𝐸𝐵𝐼𝑇𝐷𝐴) 𝐸(𝐸𝐵𝐼𝑇𝐷𝐴) – 𝑁𝐷!– 𝐸(𝑁𝐷) 𝐸(𝑁𝐷) LIQ Liquidity 𝐶𝐴!– 𝐸(𝐶𝐴) 𝐸(𝐶𝐴) – 𝐶𝐿!– 𝐸 𝐶𝐿 𝐸(𝐶𝐿) Control Variables
△ EPS! Change in Earning 𝐸𝑃𝑆!– 𝐸𝑃𝑆!–!
𝑃!–!
BETA Beta 𝐶𝑜𝑣(𝑇𝑅!, 𝑇𝑅!)
𝑉𝑎𝑟(𝑇𝑅!)
MCAP Market Capitalization ln (MV)
BM Book-to-Market 1
M/B
Deflating by share price rather than EPS solves the possible problem of negative earnings as denominator. The INV signal is finished goods when available, total inventory otherwise.
3.3 Method
I perform two empirical investigations to test my research hypotheses. First, I investigate the effect of fundamental signals on changes in future earnings. Abarbanell and Bushee (1997) findings show that in a linear earnings forecasting model, fundamental signals have added incremental explanatory power over current earnings changes. Following their approach3, I run the following regression on cross-‐sectional
data: △ 𝐸𝑃𝑆!!!,!= 𝑎!+ ß!𝑆!"# ! !!! + ß!"△ 𝐸𝑃𝑆!,! + 𝑢! ( (1) Where,
𝑆!"#: are the fundamental signals 𝑗 presented in table 5 of firm 𝑖 in year 𝑡.
Second, I examine the relation between fundamental signals and future abnormal returns. A significant relation between fundamental signals and future earnings does not automatically imply that abnormal returns are earned. If markets immediately incorporated the predictive nature of fundamental signals into security prices after disclosure of financial statement, no abnormal returns could be earned. Starting the research from the second hypothesis would be unreasonable as we would not know if there is a relationship between fundamental signals and future earnings on which the second hypothesis rely. Moreover, I test the relation between fundamental signals and long-‐term future abnormal returns to examine whether returns to signals diminish. This should serve as an additional evidence whether returns to signals are based on risk-‐ based explanations or earnings surprise reasons. Due to the fact that probably not all fundamental signals will predict future earnings, I also test for abnormal returns by using only fundamental signals that have a statistically significant effect on future earnings.
To examine this hypothesis, I follow the methodology of Swanson et al. (2003), which is based on Fama and Macbeth (1973) zero-‐investment portfolios. First, abnormal returns are defined as the 12-‐month total return of a firm deducted by the 12-‐month index total return, where total returns are defined as stock or index returns including reinvested dividends. I take investment positions 4 months after the end of the financial year, ensuring that all relevant information are available. Second, each fundamental
3Pooled cross-‐sectional regression is also the method in other related literature (e.g. Lev and Thiagarajan
16
signal and control variable is ranked into deciles from 0 (lowest variable values) to 9 (highest variable values) and each decile rank is divided by 9, creating variables with a range from 0 to 14. That is, equally weighted long and short positions are taken in
highest and lowest decile firms of each signal, implying zero net investment. The decile ranks are recalculated each year to incorporate for changes in portfolio positions. Abnormal returns to this method is evaluated using following cross-‐sectional OLS regression: 𝑅!!!,! = 𝑎!+ ß!𝑅𝑆!"# ! !!! + ß!" 𝑅𝐵𝐸𝑇𝐴!"+ ß!! 𝑅𝑀𝐶𝐴𝑃!"+ ß!" 𝑅𝐵𝑀!" + ß!" 𝑅 △ 𝐸𝑃𝑆!"+ 𝑢! ( (2)
Where 𝑅 in the independent variable marks the scaled decile rank.
This method simplifies the interpretation of the findings as the coefficients of equation (2) reveal the abnormal returns from a zero-‐investment portfolio contained in each signal, where a zero coefficient mean that fundamental signal is fully incorporated into market prices. The sum of all fundamental signal coefficients represents the net portfolio return from exploiting the information contained in all the signals, while controlling for other factors (Swanson et al. 2003).
Fama and Macbeth (1973) zero-‐investment portfolios raise implementation issues, as the trading strategy used in this research is hardly possible to implement. This is due to the research purpose of this paper. A simplified zero-‐investment portfolios can be formed that enables practical implementation. Another issue is that there are sometimes restrictions to take short positions and it cannot be ensured that somebody will lend the respective security on the date of the portfolio formation, which is especially crucial for illiquid stocks. Abardanell and Bushee (1998) note that transaction costs for a zero-‐investment strategy need not to be costly as only a single trade is necessary in a given security each year. Also additional trades in following years need not be executed if a firm remains in the same portfolio position. Another factor that could mitigate abnormal returns is the impact of price pressure from heavy trading, especially effecting small firms. This could have substantial effect on bid-‐ask spreads and, in doing so, could diminish the returns. However, Mohanram (2005) demonstrate
4 The firm in the lowest (highest) decile gets a value of 0 (1), the firm in the second lowest (highest) decile
the robustness of such an investment strategy by portioning his sample in several ways to handle issues related to implementation.
Due to the fact that pooled regressions bear potential cross-‐sectional correlation in the residuals, I follow Fama and Macbeth (1973) by calculating the t-‐statistic. Even though the number of regressions in my research (12 regressions) seems to be small, it is not uncommon to use this method with a similar magnitude of regressions in related literature (e.g. Lev and Thiagarajan 1993; Abardanell and Bushee 1997, 1998; Swanson et al. 2003). The t-‐statistic is calculated as follows:
𝑡(𝑝)= 𝑝
𝜎𝑝/ 𝑛 (3)
Where,
𝑡(𝑝) is the t-‐statistic of yearly coefficients 𝑝 is the average of yearly coefficients
𝜎𝑝 is the standard deviation of yearly coefficients 𝑛 is the number of regressions
18
4. RESULTS
4.1 Relation between Fundamental Signals and Future Earnings
In this section I examine the ability of fundamental signals to predict future earnings. Table 6 summarizes the regression of one-‐year-‐ahead changes in earnings on fundamental signals. The results show that while the signals for inventory, S&A expenses and effective tax rate are as expected statistically significant positive related to one-‐year-‐ahead changes in earnings; net investment, leverage and liquidity signals have an unexpected statistically significant negative relation. Other fundamental signals do not significantly predict future earnings. The research of Abarbanell and Bushee (1997) showed that inventory, gross margin, effective tax rate and labor force productivity signals are positive and statistically significant related to one-‐year-‐ahead changes in earnings. My findings are in many points in line with the results of Abarbanell and Bushee (1997).
The statistically significant positive inventory signal supports my expectation. This is in line with the findings of Abarbanell and Bushee (1997) and the explanation of Lev and Thiagarajan (1993). This finding means that a disproportionate inventory increase suggests troubles in generating sales or write-‐offs of outdated items.
The S&A expenses signal coefficient is positive related to future earnings. This is in line with theory, as an improvement in indirect cost should be apprehended in a positive manner regarding future performance (Lev and Thiagarajan 1993). No significant relation to this signal was observed in the research of Abarbanell and Bushee (1997).
The effective tax rate signal is statistically significant positive and, in doing so, correspond with the results of Abarbanell and Bushee (1997). The increase in firm’s effective tax rate indicate that earnings will not persist at current levels, predicting better future economic performance (Abarbanell and Bushee 1998).
The leverage signal coefficient is unexpectedly negative. This finding is in opposite to my expectation and means that an increase in financial leverage can be seen as a positive signal for one-‐year-‐ahead earnings. This fining can be explained with the theory that managers, on average, undertake positive net present value projects, which normally do not immediately affect firm’s profitability. In order to undertake these projects, external fund raising is often required. Due to the fact that raising funds on the debt market is less costly than on the equity market, leverage increases immediately (Dimitrov and Jain 2008).
The liquidity signal coefficient is also unexpectedly negative, indicating that an increase in current assets in excess of current liabilities is actually bad news for one-‐ year-‐ahead changes in earnings. In other words, this signal suggests that a short-‐term orientated expansionary policy supports earnings growth (Abarbanell and Bushee 1997). This interpretation seems to overrule the initial hypothesis.
The finding that accounts receivable, gross margin and labor force productivity signals have no statistically significant effect on one-‐year-‐ahead changes in earnings causes doubt on the relevance of these fundamental signals in predicting future earnings. While this outcome is consistent with the result of Abarbanell and Bushee (1997) regarding accounts receivable signal, Abarbanell and Bushee (1997) find significant gross margin and labor force productivity signals.
Table 6: Regression of One-‐Year-‐Ahead Changes in Earnings
Full model: △ 𝐸𝑃𝑆!!!,! = 𝑎 + ! ß!𝑆!"#
!!! + ß!"△ 𝐸𝑃𝑆!,!+ 𝑢! Restricted model: △ 𝐸𝑃𝑆!!!,! = 𝑎 + ß!△ 𝐸𝑃𝑆!,!+ 𝑢!
Independent Variables
Dependent Variable INT INV AR GM S&A ETR LFP NI LEV LIQ △EPSt Adj. R2
EPSt+1 Full model 0.011 0.030 0.026 0.012 0.071 0.005 0.007 -0.059 -0.010 -0.154 -0.236 10.3% T-statistic 1.258 1.385* 0.689 0.176 1.653** 1.946** 0.392 -3.110*** -2.090** -3.119*** -5.144*** Years positive 7 (6) 8 (2) 7 (2) 8 (4) 10 (1) 10 (3) 7 (3) 2 (0) 2 (1) 1 (0) 0 (0) Years negative 5 (3) 4 (0) 5 (2) 4 (1) 2 (0) 2 (1) 5 (1) 10 (3) 10 (3) 11 (5) 12 (10) Adjusted model 0,016 0,005 0,017 0,004 0,063 0,009 0,155 -‐0,188 7.1% T-statistic 1,742* 0,176** 0,328 1,876** 3,144*** 2,831*** 2,966*** -‐4,119*** Years positive 8 (6) 8 (3) 7 (1) 10 (3) 10 (4) 9 (2) 11 (4) 1 (0) Years negative 4 (1) 4 (1) 5 (0) 2 (0) 2 (1) 3 (0) 1 (0) 11 (9) Restricted model 0.017 -0.218 5.4% T-statistic 1.829** -4.116*** Years positive 8 (9) 1 (0) Years negative 4 (9) 11 (10)
Regressions are based on 6792 underlying firm year observations between 2000 and 2011 All variables are trimmed at 1% and 99% levels
Coefficients are arithmetically averaged from 12 annual regressions T-‐statistic is calculated as in equation (3)
Statistical significance level are based on one tailed tests, where * marks a level of 0.10, ** marks a level of 0.05 and *** marks a level of 0.01 Number of positive yearly coefficients with quantity of statistically significant yearly coefficients (significance level of 0.05) in parentheses Number of negative yearly coefficients with quantity of statistically significant yearly coefficients (significance level of 0.05) in parentheses INT is the intercept of the model
The restricted model in table 6 assesses whether the fundamental signals have added incremental explanatory power over current earnings changes. The only explanatory variable in the restricted model is current earnings changes, which is also included in the full model. The considerably higher adjusted R-‐squared of 10.3% in the full model compared with 5.4% in the restricted model shows that fundamental signals add incremental explanatory power over current earnings changes to predict one-‐year-‐ ahead changes in earnings. Unreported partial F-‐tests of the explanatory power of fundamental signals show also a statistical significance in 7 out of 12 years. However, these results are exaggerated, because this model allows individual signal coefficients to take on negative values.
In order to deal with this issue, I examine the explanatory power of fundamentals signals over current changes in earnings, using the model from Abarbanell and Bushee (1997) by combining signals into an composite score. This model assigns for each firm year observation a value of 1 (0) for each positive (negative) fundamental signal, with the result that each firm year observation obtains a score between 0 and 95.
The regression of one-‐year-‐ahead changes in earnings on composite score and current earnings (reported in table 7) show that the composite score is statistically significant6. This finding demonstrates the validity of each fundamental signal in the
model to predict future earnings. This means that the whole set of fundamental signals used in this research has explanatory power to predict future earnings, and, hence, I cannot reject the first hypothesis. Moreover, this finding justifies to use the entire set of fundamental signals as a trading strategy.
Due to the fact that not all fundamental signals predict future earnings, I reexamine my model by using only fundamental signals that have a statistically significant effect on future earnings (referred to as “adjusted model”). For this purpose, I change the sign of the net investment, leverage and liquidity signals, so that for all signals a positive relation can be expected. These results are also reported in table 6. The only difference to the full model is the missing statistically significance of the S&A expenses signal. Overall, the considerably lower adjusted R-‐squared of 7.1% in the
5 9 is the maximum score as I use 9 fundamental signals in my research
6 In addition, I examine an partial F-‐tests (unreported), which shows a statistical significance in 7 out of 12
