High Book-to-Market Firms - Separating Winning Winners from Losing Winners using a Contextual Approach

(1)

High Book-to-Market Firms - Separating Winning

Winners from Losing Winners using a Contextual

Approach

Abstract

This paper examines the performance of the model constructed by Piotroski (2000) in the time period of 2000-2009 and constructs three new models based on the relative performances of the firms in the data sample. The new models are constructed on the basis of medians and quartiles of firm performances in the sample as well as medians drawn from the industrial groups present in the sample, primarily to improve returns, but also in order to screen for firms that are at risk for earnings management. These additional models yield greater returns on long position investments but show considerably worse results on short positions.

UPPSALA UNIVERSITY

Department of Business Studies December 2010

Bachelor Thesis

(2)

1. Introduction

Actors on the stock market are constantly trying to find ways to make money through different investment strategies. The strategy known as fundamental analysis is widely used in different forms by practitioners. In essence, it analyzes different aspects of a company ranging from financial information to management and macroeconomic factors. In this thesis the fundamental analysis is used to analyze certain items in the financial statements, namely key value drivers, excluding other forms. Thus by making different comparisons between various items from the balance sheet and financial reports the strategy provides information about future winners and losers.

One of the perhaps more practical and well known studies of using the fundamental analysis as an investment strategy was performed by Piotroski (2000) who formed an investment strategy aimed at high book-to-market firms based on the performance of nine different signals derived from released financial information. Each signal was given a value of 1 (0) depending on whether it was positive (negative), which was then aggregated to his F_SCORE ranging form 0-9. However the scoring was not dependant on how positive or negative the signal was; this means that a 1% difference is given the same weight as a 50% difference. We see room for improvement in this area.

We aim to alter Piotroskis (2000) fundamental analysis by making contextual comparisons in three different ways based on median and quartile calculations. Mohanram (2005) provides a quite direct method of identifying context as one criterion within the fundamental analysis for deciding which companies to invest in. He analyzed low B/M firms and used the first two digits of the companies’ Standard Industry Classification codes, SIC-codes, to identify the types of companies to invest in and calculated the median for each signal per SIC-code. He then assigned his G_SCORE points based on if the signal was above (1) or below (0) the median for that type of company. This serves as the starting point for our study.

(4)

they place themselves for every specific signal. 2 point is given to companies in the 4th quartile, 1 point to 3rd and 2nd quartile values and 0 points to 1st the quartile. This second contextual comparison we call SQ_SCORE (sample quartile score).

The third scoring system in our study is based on the different industries (defined as the first two ICB -code digits) being represented in our sample. By scoring the companies depending on whether they place themselves above or below the industry median for each signal we hope to be able to catch industry wide trends and thus only keep the winning winners. We call this score IM_SCORE, with a scoring range of 0-9.

This way we want to refine the investment strategy developed by Piotroski (2000). In addition to screen the winning winners from the perhaps losing winner we hope to be able to discard companies conducting earnings management. The reason for this is that studies have shown that management tend to manipulate earnings to reach just above zero (that other wise would have been negative) once they realise that the results are not what they had hoped for (Dechow and Schrand, 2004). Therefore we believe that by discriminating the firms that present results slightly above zero, but hopefully below the median, we will avoid companies associated with earning management.

With the aggregated scores from SM_SCORE, SQ_SCORE and IM_SCORE we form "long position" portfolios consisting of firms with high scores, "short position" portfolios consisting of firms with low scores as well as zero-investment portfolios based on both types of firms. By “long position” we mean that we buy the shares that we believe will increase its share value, by “short position” we mean that we short sell the expected losers. The zero investment strategy then aggregate the gains from the long and the short position to collect maximum returns.

(5)

Our research question for this study is: Can the fundamental analysis of high book-to-market

firms be improved by taking the context into account? If yes - which system performs best: the sample median, the sample quartiles, or the industry medians?

(6)

2. Theoretical framework

The investment strategy in our paper focus on high book-to-market firms (hereafter “high B/M firms”) analyzed through usage of the fundamental analysis, after which the results from this process is sorted by using a contextual selection. Finally we hopefully add another positive result that our contextual selection brings, namely to avoid a large part of the companies that manipulates their results by earnings management. We hypothesize that our scoring methods avoid firms that manipulate their results by earnings management, thereby improving the long positions.

2.1 The Efficient Market Hypothesis

Within the discussion of allocating abnormal returns (which in our case is done by using the fundamental analysis) one has to mention the efficient market hypothesis, EHM. Even though highly criticized for not being fully true, the market and its actors can be said to rely on the premise that the market is efficient (at least in some way). According to theory there are three different forms of market efficiencies; weak, semi-strong, and strong (Fama, 1970). In the weak form the stock prices are adjusted according to only historical information, while the prices in the semi-strong market adjust for all publicly available information. Finally the strong market means that prices are adjusted for all information no matter whether publicly available or not. Which of these forms that the market has is highly debated, the only fact that seems certain is that the market is not strong – otherwise there would have been no room for abnormal returns. Accordingly the mere fact that there is room for collecting abnormal returns cast doubt on the hypothesis of the efficient market. Despite this the hypothesis still fills an enormously important function as a rational or reference for all the actors on the market. Without this rational or reference we all would be lost, and no one would trust or believe in the market (Fama, 1970).

2.2 B/M ratios

(7)

measuring risk. By testing the relation between b and the market returns during different periods Fama and French were able to show that the relation disappeared during large periods, and thus showed evidence of that b might not be as suitable for measuring the expected return, as earlier thought. They found that B/M values (high) as well as size (small) combined capture the cross sectional variation in stock returns, thus a relationship between on the one hand expected return/risk and on the other hand B/M values and size.

On average high B/M firms, also known as value-stocks, earn significantly higher excess returns than low B/M firms, also know as glamour stocks. The reason for the low-BM stock’s frequent underperformance in relation to the high-B/M’s stock is that investors believe that the good previous performance is an indicator of good future performance. Thus the low-B/M stocks are highly associated with a risk of overvaluation. High low-B/M stocks, on the other side, have typically performed below expectations due to the fact that these companies are often financially distressed (Fama and French, 1992). One can simply say that the investors have lost faith in the future performance of the company. The fact that these firms often are financially distressed introduces implications, namely that changes in various financial statement items have different meanings than for firms that are not financially distressed. While Fama and French (1992, 1993) attribute the B/M effect for non-observed risks (as an alternative to b), Lakonishok et al. (1994) ascribe it to mispricing. They support evidence that investors, in general, are overly optimistic in glamour stocks that have strong current earnings, thus believed to continue to generate strong earnings (which is far from guaranteed). By focusing on value stocks and accordingly avoiding glamour stocks Lakonishok et. al. are able to show positive abnormal returns. They explain these returns as a consequnse of overvaluation of future cash flows, i.e misspricing.

(8)

2.3 Fundamental analysis

All companies on the market continuously release publicly available information, according to various accounting rules and laws. This information describes the company’s present situation, while trying to give a hint about the future. Since the information is so extensive it might be hard to screen which part of it that is important and meaningful, and which is not. The execution of the analysis is not complicated; rather the whole point of it is to be simple and to provide easily understandable calculations, using only key value drivers – usually by comparing changes in different items of the financial statement from year to year. This might seem quite rudimentary, but by comparing changes in a large set of different items it is possible to get a quite accurate forecast of the company’s future performance which is not fully incorporated in the market price.

One of the first articles written on the fundamental analysis was the “Financial statement

analysis and the prediction of stock returns” by Ou and Penman (1989). Ou and Penman

combined a large set of financial information from the financial statement into one single measure of the future prospect of the company, on which they based their decision of whether to invest or not. By doing this they were able to outperform the average rate of return on the market.

The idea was further developed by Lev and Thiagarajan (1993) who wanted to investigate the value relevance of the fundamental analysis by focusing on key value drivers (also called

signals). They computed the percentage change for each signal, compared to last year, and

then compared it to the percentage change in sales. In doing so they were able to get an idea of whether the company was going in the “right” direction or not; a larger increase in e.g. inventories than sales generally indicates a negative development, and therefore a declining share price can be expected (as well as the other way around). More specifically the fundamentals added approximately 70% to the explanatory power of earnings with respect to excess returns (Lev and Thiagarajan, 1993).

(9)

all publicly available information. Nine signals were chosen, and then tested separately – which meant that they got the ability to look at the reciprocal importance among the signals. Through this zero-investment strategy they were able to generate abnormal returns of 13.2% on a 12-month horizon.

2.3.1 Fundamental analysis aimed at B/M firms

In 2000 Piotroski presented a new approach to the fundamental analysis by combining the results from Ou and Penman (1989), Lev and Thiagarajan (1993), Abarbanell and Bushee, (1998) with the results from Fama and French (1992, 1993), Lakonishok et al., (1994) by creating a zero-investment strategy aimed at a high B/M firms.

Piotroski formed his analysis on 9 different signals divided into three different areas (profitability, source of funds, and efficiency). Companies getting positive (negative) signal values earned a value of 1 (0) on every signal. He then aggregated all the signal values to his

F_SCORE, ranging from 0-9. He then invests in the “winners” having aggregated signal values

of 8-9 and take a short position in the “losers” with values of 0-1. By doing this he is able to generate excess hedged returns of 23% on average over a period of 20 years and 7.5% on invested long position returns.

2.3.2 The importance of contextual analysis in using the fundamental analysis

Mohanram (2005) follows Piotroskis’ footsteps by forming another zero-investment strategy but for low-B/M firms. This article further distinguishes itself from (Piotroski, 2000) by basing the analysis on signals being suitable for the relevant context, namely low BM firms.

Mohanram forms his zero investment strategy using a G_SCORE, which is based on eight different signals aggregated to a total score. Whether or not points are given for each signal is decided by comparing it to the industry-median. The industry was defined by the first two digits of the SIC-code of each firm in the sample. However to acquire a positive score the value had to exceed the industry median, but also be greater zero in absolute numbers. We believe this is key; namely not to look at the individual scores from various companies isolated, but to compare them with each other to calculate the relative performance.

(10)

and take good positions on the short side. Mohanram further tests the F_SCORE from Piotroskis article on low BM firms as well as the G_SCORE on high B/M firms, where the results showed that the scores were not suitable for using on each other’s samples. This is great support of the importance of contextual consideration in constructing a fundamental analysis, and especially what signals to build it upon.

Even if the contextual analysis that we are going to conduct is not identical to Mohanram (2005) we believe that this article shows the importance of comparing signals within the sample in order not only to detect well-performing firms but also to catch industry-wide trends and how they affect performance.

2.4 Using medians and quartiles – lessening the effects of earnings management

(11)

(12)

3. Methodology

3.1 Data sample

Our starting sample consists of all companies on NYSE (New York Stock Exchange), NYSE– AMEX, and NASDAQ (National Association of Securities Dealers Automated Quotations), from the period of 1997-2009 available from Thomson Reuter’s service DATASTREAM. This results in 86395 firm-year observations. After removing all firms that did not have information sufficient to calculate values, we ranked the firms based on their values for every year and thereby got the top quintile of B/M firms (hereafter called

B/M-Q5). From the B/M-Q5 sample we removed the companies without the necessary

information needed to calculate signals, resulting in 1807 firm-year observations. For reasons unknown to us, there was a remarkable loss of observations when attempting to retrieve information on firms current liabilities, total assets and net sales.

It should be noted that the B/M-values used to calculate signals for year t for obvious reasons are taken from year t-1 in order to avoid hindsight bias. Companies that delist at some point in our time period are retained, until they delist, in order to avoid survivorship bias.

After a closer look at the medians, means and standard deviations we see that extreme values affect the sample. To avoid this we first discard the obvious extreme outliers (using a box plot diagram), secondly we replace the observations having values exceeding the 98th

percentile with the exact value of said percentile through a process called Winsorising. By doing this we replace observations that would have detrimental effects on the results from our scoring methods thereby retaining quality of the statistical tests. Having single observations affecting the entire sample makes it impossible to make general conclusions about the scoring systems. Accordingly this screening seems like most appropriate alternative available for us. This boils down to a final sample consisting of 1807 firm-year observations.

3.2 The reasoning behind our scoring methods

(13)

adjustments are needed however due to differences of available datasets on COMPUSTAT, used by Piotroski, and DATASTREAM, used by us. E.g. for computing the cash flow Piotroski uses cash flow from operations while we use net cash flow - operating activities. See below in section 3.3 for more specific information regarding how we have calculated the signals. Our method for achieving the results we seek is by awarding points not only if a firms signals are positive or not like Piotroski (2000), but how the signals perform in relation to other firms as well. This is done by testing three different contextual scoring systems: 1) a sample

median score, (SM_SCORE), 2) a sample quartile score (SQ_SCORE), and 3) an industry median score (IM_SCORE). These scores are hereafter abbreviated SM_, SQ_, and IM_

respectively.

3.2.1 The Sample Median and the Sample Quartiles

The SM_ is constructed by taking the median of the signals, and award the score 1 if the value is positive as well as above the median, and zero otherwise. For the SQ_ we divide the signal values into quartiles. For the companies receiving: i) signal values in the 4th_quartile,

and ii) positive absolute values – a value of 2 will be awarded. For companies receiving: i) signal values in the 2nd_{and 3}rd_{quartiles, and ii) positive absolute values – a value of 1 will be}

(14)

3.2.2 The Industry Median

The IM_ is constructed in a number of steps. The B/M Q5-sample for each year is first divided into several groups based on the first two digits of their ICB-codes1_{, a global industry}

index available from DATASTREAM. For all industries being represented within our sample we then compute the signal medians for the industries respectively. For the industry group where there are less than three companies we set the median to zero, since the median does not provide any information in this case. The companies having positive signal values above the industry median are awarded with the value of 1, otherwise 0. Through this score we expect to control for industry wide trends in addition to the reasoning behind the other scores.

3.3 The signals

This section provides a more thorough explanation of the signals are defined and computed in this study.

Return on assets (ROA) is defined as net income before extraordinary items scaled by

beginning-of-the-year total assets. This signal measures the (percentage) yield the company earns in relation to total assets at present time which is a good measure for how profitable the company is - the higher the number, the “better” performance of the company.

Cash flow from operations (CFO) is defined as cash flow from operations scaled by

beginning-of-the-year total assets. This measures the amount of cash that is generated through the company’s running business, which means that a high number is preferable.

Change in return on assets (Δ_ROA) is defined as current year’s ROA less the previous year’s

ROA. As stated earlier about the ROA, it is preferable that the company generates more cash than last year, which is why an increase (decrease) is deemed positive (negative).

Accruals (ACCRUAL) is defined as net income before extraordinary items less cash flow from

operations, scaled by beginning-of-the-year total assets. Accruals are an accounting based technicality that enables you to split costs and/or revenues over several periods. We deem an increase in accruals as negative since this often is a sign of a decreasing ability of paying costs and debts (which is why you “push them forward”).

(15)

Leverage (LEVER) is defined as long term debts scaled by average total assets, and Δ_LEVER

is defined as current year’s leverage less previous year’s leverage. The leverage gives a ratio of the long-term debts relation to the assets of the company. In general an increase in leverage does not necessarily have to be a bad sign, however since we mainly deal with financially distressed companies we assume a decrease (increase) to be positive (negative). Further, a Δ_LEVER of 0 indicates that the company has no long-term debts at all, which in financially distressed firms we deem as positive. This is due to the fact that is highly unlikely that a company retains exactly the same ratio of total debt to average total assets over one year of operation.

Liquidity (LIQUID) is defined as the current year’s ratio between current assets and current

liabilities at year-end, and Δ_Liquid is the current year’s ratio less the previous year’s. Liquidity represents the ability to pay (current) liabilities, thus an investor wants the liquidity to be as high as possibly which is why we deem an increase (decrease) as positive (negative).

Equity offer (EQ_OFFER) is deemed positive (negative) if common equity has not (has) been

issued. The reason for a company to issue common equity is to raise more external capital for running the business, this implies a need of money.

Profit margin (MARGIN) is defined as net sales less total operating expenses scaled by net

sales. Δ_MARGIN is therefore defined as the current year’s MARGIN less previous year’s MARGIN. This signal measures what part of the sales that the company gets to keep as earnings, meaning that the higher this is the better it is. Accordingly we deem an increase (decrease) as positive (negative).

Turnover ratio (TURN) is defined in our study as total sales scaled by beginning of the year’s

(16)

Signal F_SCORE SM_SCORE IM_SCORE SQ_SCORE ROA (Signal>0)=1 (Signal>0,median)=1 (Signal>0,industry median)=1 (Signal<q1)=0

(0,q1<Signal>q3)=1 (Signal>q3)=2

CFO (Signal>0)=1 (Signal>0,median)=1 (Signal>0,industry median)=1 (Signal<q1)=0 (0,q1<Signal>q3)=1 (Signal>q3)=2

Δ_ROA (Signal>0)=1 (Signal>0,median)=1 (Signal>0,industry median)=1 (Signal<q1)=0 (0,q1<Signal>q3)=1 (Signal>q3)=2

ACCRUAL (Signal<0)=1 (Signal≤0,median)=1 (Signal≤0,industry median)=1 (Signal>0,q3)=0 (0,q3>Signal>q1)=1 (Signal<q2)=2

Δ_LEVER (Signal<0)=1 (Signal≤0)=1 (Signal≤0)=1 (Signal>q3)=0

(0,q3>Signal>q1)=1 (Signal<q2)=2

Δ_LIQUID (Signal>0)=1 (Signal≥0,median)=1 (Signal≥0,industry median)=1 (Signal<q1)=0 (0,q1<Signal>q3)=1 (Signal>q3)=2

EQ_OFFER (Signal≤0)=1 (Signal≤0)=1 (Signal≤0)=1 (Signal≤0)=2

Δ_MARGIN (Signal>0)=1 (Signal>0,median)=1 (Signal>0,industry median)=1 (Signal<q1)=0 (0,q1<Signal>q3)=1 (Signal>q3)=2

Δ_TURN (Signal>0)=1 (Signal>0,median)=1 (Signal>0,industry median)=1 (Signal<q1)=0 (0,q1<Signal>q3)=1 (Signal>q3)=2

Table 3.1 - Summary of scoring mechanisms for all signals and scores

3.4 Calculations of returns

(17)

3.5 Statistical tests

In order to statistically evaluate our results we use the Fama-Macbeth type regressions (Fama and MacBeth, 1973). These tests are performed by calculating the difference of returns between the benchmark F_SCORE and the contextual scores, and the median and standard deviations of said differences. The t-stats are then computed by

(18)

4. Results

4.1 Scoring Metrics

The basic mechanics for the portfolios are quite similar but with subtle differences. Do the mechanics result in absolute differences between the aggregated scores? The following table shows the averages and the standard deviations between the scoring systems.

Year Average Std. Dev Average Std. Dev Average Std. Dev Average Std. Dev

2000 4,58 1,65 4,00 1,69 3,95 1,70 7,22 2,98 2001 4,00 1,54 3,61 1,45 3,54 1,52 6,27 2,44 2002 4,12 1,61 3,73 1,52 3,60 1,52 6,56 2,79 2003 4,51 1,76 4,08 1,69 3,81 1,61 6,90 2,89 2004 5,30 1,70 4,43 1,88 4,43 1,77 7,65 2,89 2005 5,28 1,64 4,30 1,68 4,25 1,71 7,53 2,74 2006 4,85 1,67 4,29 1,73 4,16 1,69 7,23 2,87 2007 5,13 1,72 4,45 1,80 4,35 1,80 7,66 3,00 2008 4,48 1,74 4,01 1,70 3,94 1,77 6,99 2,93 2009 4,40 1,58 3,92 1,59 3,81 1,56 6,85 2,81 All years 4,66 1,66 4,08 1,67 3,98 1,66 7,08 2,83

Table 4.1 - A summary of the averages and standard deviatans for all aggregated scores

SQ_SCORE (0-18) IM_SCORE (0-9)

SM_SCORE (0-9) F_SCORE (0-9)

As might be expected there is virtually no difference in standard deviations between F_SCORE , SM_ and IM_. The average score on the other hand is lowered by circa 0,5, which would indicate that a greater number of low scoring firms in SM_ and IM_ compared to F_SCORE can be expected. Likewise, a lower number of high scoring firms can also be expected. The underlying assumption is that the scores are normally distributed.

F_SCORE works with a high scoring range of 8-9, with the lower limit here being approximately the average score plus two standard deviations. We use the same calculation for setting the high scoring range for our SM_ and IM_, resulting in the range 7-9. The low scoring range is set to the same as F_SCORE, 0-1, for both new scores.

(19)

4.2 Market returns

The following table shows the returns from a portfolio based on the B/M Q5 for each year tested in our time period. The portfolio does not discriminate between high or low scoring firms and therefore shows the market returns to which we will adjust the returns on our portfolios. Certain macroeconomic events are shown in this table, such as the dotcom-bubble bursting and the financial crisis crippling the financial market. 2009 shows deviates from the norm showing exceptional numbers for one year investments – these tendencies are shown for all following portfolios.

Year Y1 Y2 no. of firms

2000 -0,124 0,012 179 2001 0,295 0,282 153 2002 0,023 0,185 169 2003 0,435 0,015 197 2004 0,112 0,198 164 2005 0,235 0,375 169 2006 0,260 0,102 197 2007 -0,131 -0,487 192 2008 -0,362 -0,017 135 2009 0,830 N/A 256 All Years 0,157 0,074 1807

Table 4.2 - Returns on a portfolio based on the top quintile of high book-to-market firms Book-to-Market Portfolio returns

The returns displayed for each portfolio is measured in raw returns and are not adjusted for the B/M Q5 portfolio. The reasoning behind this is primarily to compare our scoring methods to F_SCORE and not the market returns.

4.3 Results F_SCORE

The results from using Piotroskis (2000) F_SCORE are shown below. This serves as a benchmark to evaluate our developed portfolios, since our goal is to improve upon Piotroskis original method.

(20)

the two of the worst years (-37,8% and -2,8% respectively), which is in accordance with expectations. Moreover, two out of the total ten years generated negative returns. It should be noted that long position investments made in 2009 yield positive returns of ca 94% whereas a short position yields -89,5% . Naturally, a short portfolio which includes firms that increase in value is less than preferable, but these results presumably follow the economic recovery during the period 2009-2010.

Looking at the two-year returns we see that positive returns grow even higher (37,4%), which is good evidence of that the investment strategy actually finds the winners. The return on two-year investments made in 2009 are unavailable since the market value for stocks 2011-05-01 is still a future date.

Y1 Y2

Year High n. High n. Low n. Low n. Hedge Hedge

2000 -0,013 5 0,276 5 0,548 3 0,428 3 0,535 0,704 2001 -0,043 4 0,192 4 -0,335 6 -0,147 6 -0,378 0,044 2002 0,080 5 0,964 5 0,241 1 -0,252 1 0,321 0,712 2003 0,702 11 0,869 11 -0,200 5 -0,620 5 0,502 0,249 2004 0,305 17 0,757 17 0,418 2 0,364 2 0,723 1,122 2005 0,574 13 0,716 13 -0,263 2 0,055 2 0,311 0,771 2006 0,184 11 0,040 11 -0,177 3 -0,352 3 0,007 -0,312 2007 -0,221 16 -0,519 16 0,358 3 0,671 3 0,137 0,151 2008 -0,486 6 -0,147 6 0,457 9 0,076 9 -0,028 -0,071

2009 0,938 6 N/A - -0,885 7 N/A - 0,052 N/A

All years 0,202 94 0,350 88 0,012 41 0,092 34 0,218 0,374

Table 4.3 - Returns on portfolios formed by using F_SCORE F_SCORE Portfolios

Y1 Y2 Y1 Y2

4.4 Results Sample Median Score

(21)

firms. Greater number of firms presumably entails greater spreading of risk as well as increase returns on long positions, but seems to harm the short position returns.

Y1 Y2

Year High n. High n. Low n. Low n. Hedge Hedge 2000 0,054 10 -0,022 10 0,561 10 0,431 10 0,615 0,410 2001 -0,064 8 0,008 8 -0,123 8 -0,054 8 -0,187 -0,046 2002 -0,144 5 1,131 5 0,140 7 -0,349 7 -0,004 0,782 2003 1,215 12 1,376 12 -0,392 9 -0,187 9 0,823 1,189 2004 0,267 23 0,665 23 0,161 9 -0,399 9 0,428 0,266 2005 0,459 21 0,569 21 -0,120 8 -0,375 8 0,339 0,194 2006 0,311 17 0,180 17 -0,125 10 -0,115 10 0,186 0,065 2007 -0,185 29 -0,565 29 0,350 9 0,491 9 0,165 -0,075 2008 -0,449 10 -0,009 10 0,486 12 0,107 12 0,037 0,099 2009 1,237 12 N/A - -1,144 12 N/A - 0,094 N/A

All years 0,270 147 0,333 135 -0,021 94 -0,045 82 0,249 0,320

Table 4.4 - Returns on portfolios formed by using SM_SCORE SM_SCORE Portfolios

Y1 Y2 Y1 Y2

(22)

Year ROA CFO ∆ROA ACCRUAL ∆LEVER ∆LIQUID EQ_OFFER ∆MARGIN ∆TURN 2000 0,01 0,04 -0,03 -0,05 0,00 -0,07 - -0,02 -0,17 2001 -0,12 -0,06 -0,01 -0,09 0,00 -0,53 - -0,01 -0,06 2002 -0,16 -0,03 -0,02 -0,11 0,00 -0,24 - -0,04 -0,08 2003 -0,08 0,00 0,00 -0,09 0,00 -0,13 - 0,01 -0,01 2004 -0,02 0,05 -0,04 -0,08 0,00 -0,03 - -0,01 -0,04 2005 0,00 0,06 0,00 -0,07 0,00 0,05 - 0,00 0,01 2006 0,01 0,04 0,01 -0,05 0,00 -0,03 - 0,01 0,03 2007 0,02 0,05 0,00 -0,05 0,00 0,02 - 0,00 0,01 2008 0,00 0,04 -0,03 -0,05 0,00 -0,02 - -0,02 -0,03 2009 -0,02 0,05 -0,04 -0,08 0,00 -0,03 - -0,01 -0,04 Average -0,03 0,02 -0,02 -0,07 0,00 -0,10 - -0,01 -0,04

Table 4.5 - The medians of all signals in the sample across all years Signal medians across all years

4.5 Results Industry Median Score

The IM_SCORE consistently underperforms in comparison to the benchmark, except in one year long positions where the difference is (ca +5%). Short positions consistently yield negative returns and the difference is considerable (-6%). As a result the hedge portfolios also underperform in comparison to the benchmark.

On the two-year returns we see larger differences from the other scores, this is mainly due to poor performance on the short side portfolio where the return is -13,2%. This should be compared to the other portfolios, where if not negative, the returns are always close to zero. The two year-long portfolio is indeed not as good as on the other portfolios, but again the reason for the substantially lower returns on the two year-hedge should be ascribed to the short side.

(23)

Y1 Y2

Year High n. High n. Low n. Low n. Hedge Hedge

2000 -0,027 10 0,029 10 0,341 8 0,119 8 0,313 0,148 2001 0,242 8 0,109 8 -0,061 9 0,016 9 0,181 0,125 2002 -0,042 5 0,753 5 0,208 7 -0,303 7 0,165 0,451 2003 0,956 12 1,156 12 -0,380 9 -0,432 9 0,577 0,724 2004 0,272 23 0,474 23 0,095 7 -0,191 7 0,367 0,283 2005 0,494 21 0,649 21 -0,286 9 -0,554 9 0,208 0,095 2006 0,080 17 -0,035 17 -0,262 11 -0,286 11 -0,182 -0,322 2007 -0,157 29 -0,473 29 0,267 10 0,548 10 0,110 0,074 2008 -0,366 10 0,131 10 0,480 14 -0,106 14 0,114 0,025

2009 1,070 12 N/A - -0,836 12 N/A - 0,234 N/A

All years 0,252 147 0,279 135 -0,043 96 -0,1322 84 0,209 0,178 Table 4.6 - Returns on portfolios formed by using the industry medians

IM_SCORE Portfolios

Y1 Y2 Y1 Y2

(24)

(25)

4.6 Results Sample Quartile Score

The results from SQ_ follow those from our previous scores. Long positions yield greater returns compared to the benchmark (+4,8%) on one year investments but comparatively worse one two year investments (-5,3%). There are noticeably fewer firms in the long position portfolios in comparison to the benchmark as well as the other scoring systems. Not only are the returns lower, but the risk is presumably greater. Short positions consistently yield negative returns indicating that the portfolios include firms which increase in value, which once again is not preferable. One year short positions yield (-9,4%) on average with 2009 being a particularly bad year with two year positions showing even worse returns, on average -15,4%. The number of firms included is considerably greater when compared to other short portfolios perhaps indicating some flaw in the set low scoring range.

The poor performance on the short positions have considerable negative consequences on the hedge portfolios, as they underperform the benchmark with considerable margin (-6,2% and -23,1%).

Y1 Y2

Year High n. High n. Low n. Low n. Hedge Hedge 2000 -0,076 5 0,131 5 0,193 13 0,001 13 0,117 0,133 2001 -0,022 1 -0,492 1 -0,304 17 -0,249 17 -0,325 -0,741 2002 0,221 4 0,931 4 0,043 23 -0,757 23 0,265 0,174 2003 1,027 8 1,306 8 -0,405 20 -0,285 20 0,622 1,021 2004 0,425 9 0,390 9 0,085 11 -0,260 11 0,510 0,130 2005 0,738 8 0,856 8 -0,152 11 -0,471 11 0,586 0,386 2006 0,042 4 0,194 4 -0,142 21 -0,030 21 -0,100 0,164 2007 -0,140 14 -0,493 14 0,263 12 0,580 12 0,123 0,087 2008 -0,506 4 -0,153 4 0,476 17 0,088 17 -0,030 -0,065 2009 0,788 9 N/A - -0,998 24 N/A - -0,210 N/A All years 0,250 66 0,267 57 -0,094 169 -0,154 145 0,156 0,143 Table 4.8 - Returns on all portfolios formed using the sample quartiles

SQ_SCORE Portfolios

Y1 Y2 Y1 Y2

4.7 Comparisons and Statistical Tests

(26)

statistical significance that can be proven, which may in part be explained by the limited number of observations, or investment years.

Year High n. High n. High n. High n.

2000 -0,013 5 0,054 10 -0,027 10 -0,076 5 2001 -0,043 4 -0,064 8 0,242 8 -0,022 1 2002 0,080 5 -0,144 5 -0,042 5 0,221 4 2003 0,702 11 1,215 12 0,956 12 1,027 8 2004 0,305 17 0,267 23 0,272 23 0,425 9 2005 0,574 13 0,459 21 0,494 21 0,738 8 2006 0,184 11 0,311 17 0,080 17 0,042 4 2007 -0,221 16 -0,185 29 -0,157 29 -0,140 14 2008 -0,486 6 -0,449 10 -0,366 10 -0,506 4 2009 0,938 6 1,237 12 1,070 12 0,788 9 All years 0,202 94 0,270 147 0,252 147 0,250 66 t-stat 0,103 0,109 0,101

Table 4.9 - Returns on one year long position portfolios for all scores Long Y1 Comparisons

F_SCORE SM_SCORE IM_SCORE SQ_SCORE

Year High n. High n. High n. High n.

2000 0,276 5 -0,022 10 0,029 10 0,131 5 2001 0,192 4 0,008 8 0,109 8 -0,492 1 2002 0,964 5 1,131 5 0,753 5 0,931 4 2003 0,869 11 1,376 12 1,156 12 1,306 8 2004 0,757 17 0,665 23 0,474 23 0,390 9 2005 0,716 13 0,569 21 0,649 21 0,856 8 2006 0,040 11 0,180 17 -0,035 17 0,194 4 2007 -0,519 16 -0,565 29 -0,473 29 -0,493 14 2008 -0,147 6 -0,009 10 0,131 10 -0,153 4

2009 N/A - N/A - N/A - N/A

-All years 0,350 88 0,370 135 0,310 135 0,297 57

t-stat 0,028 -0,063 -0,055

Table 4.10 - Returns on two year long position portfolios for all scores Long Y2 Comparisons

(27)

Year Y1 Y2 Y1 Y2 Y1 Y2 Y1 Y2 2000 0,535 0,704 0,615 0,410 0,313 0,148 0,117 0,133 2001 -0,378 0,044 -0,187 -0,046 0,181 0,125 -0,325 -0,741 2002 0,321 0,712 -0,004 0,782 0,165 0,451 0,265 0,174 2003 0,502 0,249 0,823 1,189 0,577 0,724 0,622 1,021 2004 0,723 1,122 0,428 0,266 0,367 0,283 0,510 0,130 2005 0,311 0,771 0,339 0,194 0,208 0,095 0,586 0,386 2006 0,007 -0,312 0,186 0,065 -0,182 -0,322 -0,100 0,164 2007 0,137 0,151 0,165 -0,075 0,110 0,074 0,123 0,087 2008 -0,028 -0,071 0,037 0,099 0,114 0,025 -0,030 -0,065 2009 0,052 N/A 0,094 N/A 0,234 N/A -0,210 N/A All years 0,218 0,374 0,249 0,320 0,209 0,178 0,156 0,143

t-stat 0,049 -0,034 -0,011 -0,154 -0,099 -0,133

Table 4.11 - Returns on one and two year hedge portfolios for all scores Hedge Portfolio Comparisons

(28)

5. Analysis

Results from our empirical test are not fully as was hypothesized. Our scores seemingly perform as intended in long positions, primarily one year investments. Our portfolios yield between 5% and 7% better returns than the benchmark on one year investments – on two year investments the results are more scattered. SM_ emerges as the lead performer in this aspect showing improved returns on both the one and two year positions. We believe that our scoring methods emphasize top performers, due to the fact that a firm consistently needs above-median results in order to be assigned a value above zero. The increased scoring range of 7-9 for long investments allow for a greater number of firms to be invested in. All our modified portfolios show better results than the benchmark on one year positions, but the reason why these improved returns are not as persistent through the second year is unclear and might indicate weaknesses in forming such portfolios with numerous firms. IM_ retains some of the improvements made with SM_, but we believe the "dilution" of medians due to the industrial groups weaken the effect. The downsides however are even more prominent on the short position in comparison to the benchmark.

Short positions are consistently underperforming the benchmark, regardless of scoring system. Greater number of firms are included in the portfolios but results are increasingly negative. We believe this is due to stricter conditions for awarding points – valuable companies that would have positioned themselves in a ”non-portfolio” in F_SCORE with points between 2 and 7 are moved downwards into a short position where they accordingly harm returns. This trend is especially emphasized in our SQ_ where the mechanics for reaching high aggregated signal values are also the most demanding.

Hedge portfolios are formed by combining long and short positions in order to form a zero-investment strategy, and they are clearly on average harmed by the shortcomings of the short positions. In this case the F_SCORE shows a greater ability of generating larger returns on short positions rather than long positions in comparison to the other scores.

(29)

short positions. Unfortunately this is necessary in order to achieve a correct sorting of winners into long positions.

The various scoring methods introduced by us increase the demands for firms to be assigned points based on the fundamental signals, which in combination with the increased scoring range used to deem whether or not firms are investment worthy seemingly yield positive returns. However, we ask ourselves if the improved returns for our portfolios are due to the extended range for long positions when compared to the benchmark. Are improved returns due to greater amount of firms or due to better screening for top performers? Assuming that scores are normally distributed, the score range of 7-9 we set for long position investments would therefore contain the top 5% performers. This leads us to believe that it is indeed the scoring mechanics themselves, and not extended long position ranges, that have the aforementioned effect. This is even more prominent considering the long position range of F_SCORE where the average signal value plus two standard deviations gives the range 8-9 points.

Regarding the effects of earnings management and the screening for firms that are associated with such, the possible conclusions are highly uncertain in regards to our results. Whether or not our improved returns on long positions are due to "healthier" firms or merely better performing firms is difficult to ascertain. Not only are there difficulties of measurement but also perhaps due to the nature of earnings management which is hardly readily available in solid numbers, it is difficult to test for. These factors are perhaps reinforced by our somewhat lacking data sample due to problems mentioned in the methodology section. As the statistical tests performed on our scoring method do not show any significance it makes it even more difficult to draw any conclusions. We do however believe our scoring methods have the desired effect.

(30)

(31)

6. Conclusions

In this article we have looked at Piotroskis (2000) investment strategy aimed at firms with high B/M values using only publicly available financial information to form investment portfolios. The time period analyzed (2000-2009) yields on average hedged returns of 21,8% using F_SCORE, which should be compared to the return of the high B/M portfolio 15,7% (accordingly 6,1% return adjusted for the value-weighted market return). Hence before looking at our scores it should be said that this time period is one where the fundamental analysis performed according to Piotroskis method does not achieve quite as positive results as he did in the period of 1976-1996. We also see that the year 2009 distorts the average returns for the one-year horizon on all the scores, including Piotroskis F_SCORE, where the short positions yield approximately 100% negative returns despite having been deemed poor performers.

When constructing our different scoring methods we believed that using a contextual approach when scoring the different signals would refine the results from the F_SCORE. The way we constructed the scoring mechanics we wanted to create a system where only the real rockets among the winners (from the F_SCORE) would run through our filter for picking the companies to include in the long positions. The method of using medians further evolved into using quartiles to the same effect. The same thoughts were the source to our IM_, where we wanted to capture industry wide trends.

However these were not our only hopes, as can be seen in the theory section when reading under paragraph 2.3. Even though being far from the reason why we made the tests we did, we were hoping to be able to discard companies conducting earnings management through our SQ_– where we discriminate the companies presenting signal values in the middle of the sample within the scores. The rational behind these thoughts were the findings from Hayn (1995), Burgstahler and Dichev (1997), Dechow and Schrand (2004) where they show that management tend to manipulate negative earnings to reach slightly above zero when they fall short of forecasts.

(32)

that indicate room for improvement in evaluating the signals when forming long position portfolios considering the improved returns yielded by SM_SCORE, IM_SCORE and SQ_SCORE. Short positions however are clearly not improved by such a contextual approach, rather the opposite. Due to the stricter conditions for being deemed investment-worthy, potential winners are sorted into between long and short positions and are therefore discarded from investment. In some cases firms that increase in value are instead sorted into short position portfolio, harming returns.

A way to combine the benefits from both sides would be to use SM_SCORE, or sample medians for each signal, to form long positions and F_SCORE, a straight evaluation if the signal is above or below zero, to form short positions. Further avenues of potential research would be to further research into the effects of earnings management on financially distressed firms and returns on investments made in such firms. From this study it cannot be ascertained to which extent the increased returns on long positions are due to avoidance of "dishonest" firms or simply poor performing firms.

(33)

Bibliography

ABARBANELL, J. & BUSHEE, B. 1998. Abnormal returns to a fundamental analysis strategy. Accounting Review, 73, 19-45.

BLACK, F. 1972. Capital market equilibrium with restricted borrowing. Journal of

business, 45, 444-455.

BURGSTAHLER, D. & DICHEV, I. 1997. Earnings management to avoid earnings decreases and losses. Journal of accounting and economics, 24, 99-126.

DECHOW, P. & SCHRAND, C. 2004. Earnings quality, CFA Institute.

FAMA, E. 1970. Efficient capital markets: A review of theory and empirical work. Journal

of finance, 25, 383-417.

FAMA, E. & FRENCH, K. 1992. The cross-section of expected stock returns. Journal of

finance, 47, 427-465.

FAMA, E. & FRENCH, K. 1993. Common risk factors in the returns on stocks and bonds* 1. Journal of financial economics, 33, 3-56.

FAMA, E. & MACBETH, J. 1973. Risk, return, and equilibrium: Empirical tests. The Journal

of Political Economy, 81, 607-636.

HAYN, C. 1995. The information content of losses. Journal of Accounting and Economics, 20, 125-153.

LAKONISHOK, J., SHLEIFER, A. & VISHNY, R. 1994. Contrarian investment, extrapolation, and risk. Journal of finance, 49, 1541-1578.

LEV, B. & THIAGARAJAN, S. 1993. Fundamental information analysis. Journal of

Accounting research, 31, 190-215.

LINTNER, J. 1965. The valuation of risk assets and the selection of risky investments in stock portfolios and capital budgets. The Review of Economics and Statistics, 13-37.

MOHANRAM, P. 2005. Separating Winners from Losers among LowBook-to-Market Stocks using Financial Statement Analysis. Review of Accounting Studies, 10, 133-170.

OU, J. & PENMAN, S. 1989. Financial statement analysis and the prediction of stock returns. Journal of Accounting and Economics, 11, 295-329.

PIOTROSKI, J. 2000. Value investing: The use of historical financial statement information to separate winners from losers. Journal of Accounting Research, 38, 1-41.

High Book-to-Market Firms - Separating Winning Winners from Losing Winners using a Contextual Approach