Predictors of shareholders loss : declining earnings and analysts’ forecast errors

(1)

i

Predictors of Shareholders Loss: Declining Earnings & Analysts’ Forecast Errors

Master Thesis

MSc Business Economics: Track Finance

July, 2015

Author:

Supervisor:

Milena Levicharova

Dhr. Dr. Tomislav Ladika

Abstract

Before 2005-2006 employee stock options were disclosed in footnotes, however, under FAS123R, they had to be recognized as expense, reducing the firms’ immediate accounting performance. The present paper finds that when the stock options were firstly expensed, stock analysts reacted to the firms’ declining bottom line earnings without analyzing what was driving them and downgraded their forecasts and recommendations. Investors, further misled by the analysts also reacted similarly and proceeded with unwinding their stock holdings. The actions of both parties resulted in stock return decreases. To establish the effect of the standard, the analysis constructs an instrument refracting on the exogenous variation in the mandatory compliance timing of FAS123R – when the companies had to comply was staggered depending on their fiscal year end months. The instrumental variable estimates show that the drop in accounting performance resulted in more than 26% reduction of stock returns. Analysts forecast errors increased by more than 50% and in turn resulted in reduction of total shareholder returns of 18%

(2)

Statement of Originality

This document is written by Milena Levicharova who declares to take full responsibility for the contents of this document:

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

06.07.2015

(3)

Acknowledgements

I would hereby like to establish my gratitude for the reference data, assistance,

guidance and help I was offered by my supervisor, Dhr. Dr. Tomislav Ladika, who

mentored every step of the completion of this paper. Dr. Ladika met every inquiry

of mine with a prompt, timely and diligent reply and provided me with invaluable

advice.

(4)

`

1. Introduction

In many cases, a change in policy has spill over effects which are difficult to envision in advance. Such is the case of FAS123R which mandates employee stock option expensing for the first time in the period 2005-2006. Under the accounting standard, a single line item changes place from the footnotes of the financial statements to the expenses section, seemingly with little to no implications – the information was known & nothing new is reported. American companies used to disclose the value of their stock options, yet only in a footnote. Under FAS123R they had to recognize the expense. This largely meant reporting lower earnings and presented a market signal of declining profitability. The latter was picked up by stock analysts & communicated to investors. Would they both react to the bottom line earnings only, or would they study the details of firms’ disclosures and understand that nothing is in fact changing? Would analysts get confused in their forecasts or recommendations? If they substantially reduce them, how are the investor going to react when faced with the downward revisions?

Considering that analysts are the middle man between a company and investors, the latter are likely to react to the changes in analysts’ forecasts, potentially unwinding part of their holdings and thus bidding down stock prices. This line of actions is what the present paper ventures to investigate & thus the main analysis falls into a controversial area of research without a unified verdict: the effect of stock option expensing, mandated in 2005-2006 by FAS123R, on total shareholder returns. The current research fills a gap in this respect: it hypothesizes & delves into the economic magnitude of the changes in stock returns attributable to FAS123R & translates them into dollar value.

Theory already suggest that Cash is King, i.e. that Ebitda would provide the widest and clearest picture of the magnitude of the standard’s effects. Many would argue that the true company health and abilities would only reveal themselves in the Operating Profits, since they are the ones actually generated by the companies' operations. Net Income would in turn also react and is a widely adopted performance measure. So not only is stock option expensing going

(6)

2 to influence companies’ earnings, but it also has the potential to take aback the analysts who would have to make a statement on the companies’ performance. This delineates the venue for FAS123R to affect Stock Returns: it causes a decline in firms’ earnings which without a careful inspection by the analysts can translate in downward revisions of their forecasts and recommendations. The investors as well when seeing lower profits and downward forecasts could think that the companies in their portfolios are undergoing distress and would likely reduce their holdings in such stocks to prevent losses, thus causing a reduction in stock prices.

In support, theory suggests that analysts are indeed capable of influencing stock prices, Renfro, (2015), even in the simplest of cases – e.g. before earnings announcements (Park & Stice, 2000). The addition of an expense item should logically be reflected in the operating and net income of any company, but this would also directly impact the most widely used analyst’ estimate: the EPS value. Nevertheless, the actual performance of companies does not change, especially since all the information was previously disclosed. What is expected is that some analysts could have envisioned that, others could have just relied on the bottom lines seeing only the fringe reduction in profitability. This behaviour is thus a plausible birth of a forecast error & coupled with declining profitability & the analysts’ influence, is tantamount to a (potential) conflict with shareholders’ interests. The main contribution of this paper is to show that the decrease of Ebitda & Net Income & the increase in analyst forecast errors, caused by the change in accounting rules, i.e. FAS123R, resulted in a reduction of total shareholder returns.

Having introduced the Income Statement effects of FAS123R & the analysts’ plausible confusion, two approaches are thus delineated for examining the reaction of returns to the accounting change. The first is through the decrease of the firms’ Ebitda & Net Income caused by FAS123R increasing compensation expenses. The second, instruments the Forecast Error and shows the error’s effect on stock returns. Such an analysis thus permits a crude comparison between effects magnitudes & raises further research issues, like: if analysts’ forecasts get so erroneous and able to influence stock performance, than are the financial statements capable of causing an even worse picture?

(7)

3 These two approaches lead to a natural segregation of the present analysis: (1) effect of FAS123R on stock returns through the decline in accounting earnings & (2) effect of FAS123R on stock returns through analysts’ forecasts errors when predicting earnings. These would all be analysed with the help of an instrument, so constructed as to represent the exact quarter in which companies complied with FAS123R.

Considering the introduced hypothesized inefficiencies, the details of the events interlinked with FAS123R also have a connection with the Efficient Market Hypothesis. The latter would suggest that if there was new information on the market, and the market was efficient, the new information would immediately be incorporated in share prices and no speculation could arise therefrom. In the case of FAS123R, however, there is no new information and yet, the chain reaction on the market to the accounting change is significant for many participants. Thus it could be argued that it leads to a temporary market inefficiency, adding to Maheshwari and Dhankar’s (2014) review of market overreactions for the past three decades.

A further secondary aspect touches upon another disputed area of research: the discrepancy between forecasts and recommendation. Analyzes shows that often times there is a difference between an analyst forecast and the recommendation he makes: while a sell side analyst can make an optimistic recommendation this would not always be reflected in the forecast he makes (Malmendier & Shanthikumar, 2007; Mikhail, Walther & Willis, 2007). On one hand side, a suggested explanation is that the more sophisticated measure is oriented toward more sophisticated investors: like e.g. institutional; while the recommendations target the smaller investors, lacking the resources of time and expertise to go through all forecasts. In this line of thought, the present empirical research also contributes by showing that in the vicinity of adopting accounting changes, and FAS123R in particular, forecast errors indeed translate into losses, but the downwardly revised recommendations do not result in stock return changes.

To proceed further, another idea should be introduced: that technological firms potentially suffer the most from the erroneous forecasts. A natural reasoning as to why this is so, is that these companies rely more heavily on stock option compensation than other – in many

(8)

4 cases, and in their early and growth years, their funds are mostly reverted to e.g. research and development and investing; the compensation with stocks is thus less contingent on cash availability and unwinds resources to promote their growth. Resultantly, the fair value of employee stock options can be higher for technological firms. It can be therefore hypothesized that their stock returns would be more impaired due to the acceptance of FAS123R. The present paper, however, disproves this hypothesis, although the particular finding could be due to the smaller sample of technological firms. Eventually, however, they are indeed negatively impacted, though less than the whole universe of firms analyzed.

Another secondary aspect to mention is the potential for managers to intervene and attempt to moderate effects on EPS to keep the firm’s image high. Thus, however on one hand analysts, the middlemen for investors, cannot consistently foresee which firms are optimising Net Income to avoid reporting declines (Burgstahler & Eames, 2003) & in a way managers might so further contribute to an erroneous forecast estimate. Additionally, Goldman & Kohnlback (2015) suggest that in the vicinity of FAS123R stock repurchases could have increased exactly to moderate earnings per share. Stock repurchases, however, are not only means of controlling EPS, but are part of investors’ total returns. Thus taking the authors idea one step further, it can be tested if adjusting stock returns for repurchases really moderates shareholder losses. Hence, two types of returns are hereby considered: (1) returns, adjusted for dividends, and (2) returns, adjusted for dividends and share buybacks. The paper finds that repurchases did play a role for preserving stock returns.

Thus the whole chain reaction of lowered performance, triggering erroneous forecasts, leading to investors selling their holdings, forms the basis for analysing the accounting returns, analysts’ errors & recommendations & whether the above mentioned further translate, in 2SLS, in stock return declines.

The current paper proceeds with the following parts: Section 2 provides a literature review and background on FAS123R; Section 3 presents the empirical methodology employed for

(9)

5 the analysis; Section 4 discusses the results of the research; Section 5 describes some robustness checks and Section 6 summarizes and concludes.

2. Literature Review and Background on FAS123R

2.1. Literature review

The present paper contributes to a growing body of literature documenting the spill over effects of the introduction of FAS123R and its mandatory acceptance in 2005-2006. It also falls among other, not so event-specific disputed topics.

Having already introduced the effect of FAS123R on firms’ accounting profitability, it is time to indicate how this is important for investors. When they put their money in a stock they do so for a reason: to gain from the capital appreciation or to earn dividends (ignoring speculation, voting power, etc.) (Gordon, 1959). Declining earnings, however, signal that capital gains and dividends might be at stake. If a company’s profits sharply plummet, it might end up unable to return cash to its shareholders. This would likely deter some investors from holding on to such shares & they would likely preserve their portfolios by unwinding part of their holdings, thus causing a decrease of the stock price (Skinner & Sloan, 2012).

Often though, investors rely on analysts. The latter, if ignorant for the acceptance and mechanics of FAS123R, might too downwardly revise their forecasts and recommendations, further amplifying the potential shareholder losses, by triggering even more investors to act on the lowered earnings. Hence, if some analysts allow for the stock option expensing and other do not, than it can be reasonably expected that their forecasts would be less on par with reality. As Diether, Mallow & Scherbina (2002) & Johnson (2005) prove that greater forecast dispersion is negatively correlated with stock returns, in the case of FAS123R, it can too be expected to happen. In this line of reasoning, the present research visualizes the chain reaction to FAS123R – the earnings decline, analysts get confused and downgrade stocks, investors react by selling them

(10)

6 as a consequence, the cumulative effect of FAS123R reveals itself in the stock value. Similar to the post-analyst forecast revision price drift (Chen et all., 2013), yet more exacerbated in the case of FAS123R, the present paper finds that the share prices plunged post the erroneous forecasts. The current research also builds upon Barth, Gow & Taylor (2012) who too review the spill over effects of FAS123R. Following the authors reasoning, the hereby documented increase in analyst forecast errors could be partly attributed to: (1) some managers intervening in the reported net earnings to state them higher, (2) others simply opportunistically not reporting the employee stock option expense when they were supposed to, both occurrences documented by the authors. If the managers were acting in this manner, analysts could not have consistently envisioned who are going to fall in either category .More importantly, the authors further establish that the analysts themselves too sometimes decided not to account for the value of the stock options in their forecasts, even though their estimates were thus less in line with the long term intrinsic value of the firms. The current research takes Barth, Gow & Taylor’s discovery a step further – the authors rightfully diagnosed the situation as even more confounded due to the opportunistic behaviour exhibited by both players, but the current paper translates this into a monetary loss of firm value.

Another controversial area of literature deals with the analysts’ ability to generally react efficiently. Authors like Busse & Green (2002) & Agrawal & Chen (2008) put forward results which are in support of analysts abilities, but when it comes down to accounting changes or firm peculiarities, many papers find the contrary (Dechow, Hutton & Sloan, 2010; Burgstahler & Eames, 2008; Baron at all, 2002; Abarbanell, Lanen & Verrechhia, 1995) . The present one contributes to the latter, shedding extra light on the amount of forecast error and its consequences. What is more, it exemplifies the following inefficiency: analysts sell or buy recommendations plummet right after the FAS123R acceptance. Considering the fact that the information for stock option compensations was widely available and disclosed in advance, this can be viewed in one of the following manners: either the analyst were inefficient and overoptimistic before the accounting change, or they overreacted afterwards.

(11)

7 The induced change in recommendations was in no way dew or expected, but the suggested over-optimism prior 2005-2006 compliance with FAS123R also does not find much support. It falls in the arbitrary area of investors-analysts-underwriters conflict of interest. It has widely been discussed whose interest do analysts serve, with even substantial legislation governing their relationships. The notion that sell-side analysts are always prone to give higher recommendation has too been much debated, with empirical support for each stanza. While many researches vouch for the presence of biases (Lim, 2001; Firth at all, 2012, Michaelly & Womack, 1999), Clarke at all. (2006) for example, find the complete lack of bias in the analyst recommendations, when the market is in wide unrest. What is more, the authors find that recommendation reversals do not appear information-intensive for the market: they find no causal relationship between a change in a recommendation and stock returns afterwards. The current paper relates to Clarke at all. (2006) rather straightforwardly: it finds no significant market reaction to the recommendation drops exhibited in the FAS123R periods, neither does it find upwards bias, rather the contrary in terms of recommendations. Such is, however, present in the analysts’ forecasts, prone to overstating EPS expectations.

Balsam, Reitenga, and Yin (2008) further add that many companies accelerated option vesting prior to the effect of FAS123R just to avoid reporting lower EPS. Likely, in the mandatory compliance period managers also attempted to cushion (e.g. with increased stock repurchases) the effect of the stock option expense, to moderate EPS. As analysts would highly rely on the bottom line earnings (Burgstahler & Eames, 2003; Liu, Nissim & Thomas, 2007) they would thus get even more confused in their estimates.

A side note concerns the debate on which financial metric is more reliable and more information intensive: the Ebitda or the Net Income. Many papers conclude that the former is less susceptible to tampering and more value-relevant for the firm (Black, 1998 & Liu, Nissim & Thomas, 2007, Kwon, 2009). Hence, the current paper juxtaposes the effect of FAS123R on Ebitda and Net Income, however finding both to be significantly affected and with a similar magnitude.

(12)

8 The above paragraphs introduce the main lines of contribution this paper has to the literature: (1) it analyses how stock returns react to declining accounting performance of firms, even when dropping earnings are not information intensive; (2) it shows how the acceptance of FAS123R affects analysts’ forecasts & recommendations; (3) it delves into how analysts mislead investors and also contribute to a stock price decline. The following lines are dedicated to presenting a detailed background on the functionality of FAS123R.

2.2. Background on the adoption of FAS123R

Since the essence of this paper revolves explicitly around the mandatory compliance with FAS123R, several details about the accounting change are hereby introduced. Detailed diagram of the events preceding and following the acceptance of the standard can be found in Ladika & Soutner (2014). Further extensive information on the issue can be found in the papers of Chouhary, Rajgopal, Venkatachalam (2009) & Murphy (2013).

Initially, the accounting treatment of employee stock options depended on the options intrinsic value: if the options allowed employees to buy stocks at the same price at which they were trading (the options were at-the-money) when the options were granted, virtually no expense was generated, and thus, no expense had to be reported. This was underlaid in early 1972 by the then known Accounting Principles Board, who issued Opinion 25 on the matter (Chouhary, Rajgopal, Venkatachalam, 2009).

Shortly after, the same board began considering changes in the standard governing the stock option compensation schemes. Their second proposition, requiring the fair-value of the options, at the date when they were granted, to be included in the companies’ financial statements as an expense was presented in 1993 (Ladika and Sautner, 2014). The authors proceed with describing the corporate unrest and opposition outburst which elucidated therefrom. Such a change would highly impact many market participants, especially the technological industry, which was highly intensive in relying on this sort of compensation. Since these were the booming years of the industry, it was powerful enough to overstep the

(13)

9 Accounting Principles Board and find governmental support for overruling the Boards' proposal. This was an act permitting the proliferation of many then small, now giants, technological firms, who would have experienced difficulties maintaining positive income statements – they were relying on stock option compensation mainly because of their restricted resources.

Experiencing this forceful opposition, the Accounting Principles Board supported the continuation of the “no expensing”. However, it soon started to promote the fair-value accounting and demanded the explicit disclosure of the employee stock option compensation values at least in the footnotes of the companies’ financial statements. As Ladika and Soutner (2014) point, most of the firms continued to rely on the previous version of the standard, APB Opinion 25. With the course of the years, the down of the new millennium brought about the burst of the technological bubble and corporate scandals of unseen extend (take the fall of Enron and many others). Top management of huge corporations was heavily incentivized with millions of dollars’ worth of stock options, present only in the footnotes. This unlocked a new attempt of the FASB to push for recognition of the expense. Eventually, FAS123R was enforced on December 14, 2004 (Jochem, Ladika & Soutner, 2015). Companies had to expense stock options at their fair value in their firstly due financial statements, the quarterly 10-Q reports, the authors clarify. Since the actual implementation of the standard was delayed due to the increased accounting workloads, this had a very valuable implication for the present research. FAS123R had to be adopted by firms depending on their fiscal year ends. For example, firms with fiscal year ending in June, 2005, had to start expensing the stock options from the following month – July, 2005. However, firms with fiscal year end May, 2005, had to comply in June 2006 (Choudhary, Rahgopal & Venkatachalam, 2009).

For the purposes of the present analysis, another implication of the standard finds its necessary mention: some firms voluntarily switched to fair-value stock option expensing before the actual mandatory compliance period was due; partly they accelerated option vesting in advance (Jochem, Ladika and Soutner, 2015). The present research excludes firms who did this, to remain with a sample of companies, for whom the FAS123R presented an exogenous shock

(14)

10 and was randomly complied with on the basis of their fiscal year ends. The latter were determined much in advance. Companies which changed fiscal year ends shortly prior to the acceptance of the standard, probably to benefit from the opportunity to delay stock option expensing with a year, are also excluded from the research1_.

3. Hypotheses, Methodology & Data

3.1. Hypotheses

The present research sets off to analyze several hypotheses naturally born in the circumstances of FAS123R which have not yet been tested in the scientific literature. The previous paragraphs have already broadly introduced them, namely:

H1:FAS123R lead to a significant decrease in firms’ accounting performance, which resulted in

declining firm value

H2: FAS123R led to increased forecast errors, which also resulted in loss of shareholder value

The above mentioned hypotheses are the core ones for the present research, the first being central for the analysis and expected to reveal in full the effect of the accounting change. The second shows the role of analysts to the loss of shareholder value.

Two other hypotheses, secondary to the previous ones, also find proof in this paper:

H3: FAS123R resulted in lower buy-sell recommendations, which did not cause a drop in stock

returns

H4: The returns adjusted for dividends and share repurchases, as compared to returns adjusted

only for dividends, exhibited a lesser effect of FAS123R.

(15)

11 As already previously introduced, there might exist a discrepancy between analysts’ forecasts and their recommendations. Hypothesis 3 therefore also looks into the latter two and their effect on stock returns. In addition, knowing if and how much the share buybacks really contributed to the preservation of stock returns is likely to interest firms, investors and scientist and is hereby tested in a time, when share repurchases were in a way needed. The following lines describe the identification of the instrument and the methodology for testing the hypotheses.

3.2. The instrument - Identification strategy

In the present case, the natural setting of FAS123R provides an opportunity to device an exogenous instrument which can help establishing a relationship beyond collinear between the stock option expensing and the suggested outcome variables. The current research takes advantage of the staggered, randomized compliance of firms with FAS123R. The exact quarter of compliance depended on companies’ fiscal year end months, which were determined unparalleled with the event of mandating the compliance with the new accounting change. A firm with fiscal quarter ending in June, could thus experience a change in Ebitda, Net Income, analyst forecast and recommendation in the following quarter. Thus in the quarter post the mandation of FAS123R, companies’ stock returns could react adversely to the earnings measures and analysts opinions. These expectations pertain to the above listed hypotheses.

Further on the identification strategy: the random variation of firms’ fiscal year ends (and thus compliance with the accounting change) permit the identification of FAS123R as an exogenous instrument for the hypothesized plunge in stock returns. With the help of fiscal year end months then, two dummy variables are designed to serve as instruments.

The first one if FAS123R Immediately upon acceptance which is a dummy variable equal to 1 for the exact quarter when the firms have to comply with the standard, designated as [t=0]. The variable equals 0 in all previous quarters. FAS123R Immediately upon acceptance permits the immediate reaction to the standard to be established, in the quarter exactly following the acceptance of FAS123R. The highest number of firms complying with the standard in a single

(16)

12 quarter is approximately 1700 in the 4th_{quarter of 2006. The availability of quarterly control}

variable values, however, constrains the total number of observations to about the same number and thus faults the analysis. As a consequence, tests with this dummy are abandoned & a second dummy is devised & is the central for this paper, FAS123R Before and After.

FAS123R Before and After takes the value 1 in all quarters post the acceptance of the

standard and the value of 0 in the time preceding the event. Thus, a long term, before and after, Local Average treatment effect can be established for the whole data sample. The tests of the hypotheses and the identification strategy are therefore implemented with the use of two stage least squares regressions for firm i and calendar quarter t+1. The variables used throughout the research are grouped in outcome variables, main variables of interest and controls. Hence, the following paragraphs introduce each of the variable groups, clarifying the predominant and common ones for each regression specification.

3.3. Outcome Variables

As the above mentioned hypotheses suggest, the outcome variables are the following: Ebitda & Net income levels, EPS forecast errors, analyst buy-sell recommendations & stock returns. Each one is taken at [t+1], indicating that its first occurrence is considered after the acceptance of FAS123R. Ebitda and Net Income levels are percentage values of the ratio of the earnings variable taken relative to the companies’ assets, thus standardizing their values. All regression specifications involving the Ebitda and Net income consider these values at time [t+1], when FAS123R takes effect at time [t=0].

However, the Forecasts issued for Net Income [t+1], are in fact issued one quarter in advance, in time [t=0]. Therefore, the present paper analyzes Forecast Errors & Analyst Recommendation at time [t=0]2_{. The Forecast Error concerns the consensus EPS estimate}

2_{The correct quarter of the forecasts was identified with the help of data item “fpi” from}

(17)

13 submitted by analysts. In particular, it is measured as the difference between the average (or median) estimate and the actual reported value of EPS (occurring one quarter after the release of the forecasts). The analysts’ recommendations graduate from a “strong sell” to a “strong buy” which are translated into numbers ranging from 1 to 5. Since IBES only has monthly recommendations, these are averaged per quarter.

Stock Returns are calculated with adjusted close stock prices, allowing for dividends. In addition, a second Stock Returns variable is constructed and it represents the latter value further adjusted for share repurchases. The current paper expects that each outcome variable is negatively impacted by FAS123R & that the refined TSR value, allowing for the stock repurchases should exhibit a lesser change as contrasted to the TSR only adjusted for dividends.

3.4. Main variables of interest

As already introduced, the central exogenous variable employed in the analysis is the

FAS123R before and after 0/1 dummy. It is shown in the first stage results of several regression

constructs & it instruments the Ebitda, Net Income & Analysts’ Forecast Error in a two stage least squares model affecting Stock Returns. The regressions of Returns on the instrumented Ebitda & Net Income exemplify the effect of the standard most directly, since Returns are more correlated with the profitability values of a firm than with the opinions of the analysts covering it. However, the regressions find significance in almost all specifications (a 2SLS regression focusing on the Recommendations, instrumented with the dummy, shows no significant effect of their change on the stock returns).

equal to 7, to be the one of interest to all researches and the one actually involved in practice, e.g. Delavigna & Polet, (2009) also use these values.

(18)

14

3.5. Firm specific control variables

The control variables employed in the regressions are mainly firm specific and have been shown significantly relevant in related previous researches. All of them are quarterly data & the observations in each regression are limited by its availability. When all outcome variables are taken at time [t+1], all control variables are taken in the quarter before [t=0] to avoid significant collinearity and technical dependency between the variables. For Forecast Errors, which are taken at [t=0], regressors are taken at [t-1]. The inclusion of additional lagged values of some of the controls is also exemplified in several specifications. The lagged values, however, exhibit diminishing significance, the farther back in time they are taken. An example of controls present in all specifications is Total Assets of firms, as indicator of their size. This as well as companies’

Book-to-market ratio, are found to convey information on e.g. coverage by number of analysts.

The consensus forecasts for these firms are often more accurate, because of the increased coverage and analyst herding behavior: the bigger companies are covered by a greater number of analysts (Previts at all., 1994). Another variable controlled for is Stock returns. When this is not an outcome variable, it is presented as a control one, explaining part of the variation in the dependent variable.

Additional performance measures of the companies are also included: like their

Sales-to-Assets level or the level of indebtedness, captured in Debt-to-Sales-to-Assets values. The market model

additionally incorporates the Market premium as a regressor. This variable is directly taken from the Fama and French library and represents the Return on the US Market portfolio less the assumed risk free rate for the United States. For most of the regression specifications Stock

Option Grant Value is a significant explanatory variable. Indeed it is highly impacted by the

accounting change introduced by FAS123R but no collinearity is exhibited especially since a time gap of at least one quarter between the outcome and the dependent variables is envisioned.

Further, when analyzing Forecast Errors, the regressions control for the EPS actual value of the prior quarter. Similarly, direct dependency between the variables is avoided through the use of lagged values only. The single control which is taken in the same period as the outcome

(19)

15 variable is the Market Premium. They are both considered in time [t+1]. Industries are identified with the help of the extended Fama and French’s classification of 48 classes. Controlling for industry fixed effects proves significant in each of the regressions and explains a large part of the outcomes’ variation. The same can be said for year and quarter fixed effects. The former are present in each of the specifications and often found to significantly reduce the importance of the main variable of interest, i.e. they too have significant explanatory power. In some of the regressions, however, the quarterly fixed effects are found highly collinear with the Market

Premium and are sometimes omitted, in its favour. This is identified in each table of results. The

following lines introduce the main equations estimated throughout the research.

3.6. Methodology

3.6.1. Main hypotheses

The most significant finding of the paper lies in testing the hypothesis that company profitability was impaired with the acceptance of the FAS123R, and this translated into a loss of shareholder value. To test for this, the methodology starts with a first stage regression of the profitability metrics on the instrument FAS123 before and after.

(1)𝐸𝑏𝑖𝑡𝑑𝑎_𝐿𝑒𝑣𝑒𝑙_𝑖,𝑡+1= 𝛾₁∗ 𝐹𝐴𝑆 𝑏𝑒𝑓&𝑎𝑓𝑡𝑒𝑟_𝑖,𝑡+ 𝛾_𝑞∗ 𝑋_{𝑖,𝑡−1}+ 𝛾_𝑞∗ 𝑋_𝑖,𝑡 + 𝛾_𝑞∗ 𝑋_{𝑖,𝑡−1}+ λ_𝑡+ 𝜈𝑡+ 𝜇𝑡+ 𝜀𝑡

(2)𝑁𝑒𝑡𝐼𝑛𝑐𝑜𝑚𝑒_𝐿𝑒𝑣𝑒𝑙_𝑖,𝑡+1

= 𝛾₂ ∗ 𝐹𝐴𝑆𝑏𝑒𝑓&𝑎𝑓𝑡𝑒𝑟_𝑖,𝑡+ 𝛾_𝑞∗ 𝑋_{𝑖,𝑡−1}+ 𝛾_𝑚∗ 𝑋_𝑖,𝑡 + 𝛾_𝑚∗ 𝑋_{𝑖,𝑡−1}+ 𝜇_𝑡+ λ_𝑡+ 𝜈_𝑡 + 𝜀_𝑡

The Outcome variables as already introduced, are the respective profitability measures, standardized to the firm’s total assets: in regression (1) this is the Ebitda level, in equation (2) the Net Income level. The right hand side variables are the same in all specifications and contain company specific controls, time fixed effects, industry fixed effects and an error term. The coefficient of interest is the one of the instrument 𝐹𝐴𝑆𝑏𝑒𝑓&𝑎𝑓𝑡𝑒𝑟𝑖,𝑡, i.e. 𝛾1 𝑜𝑟 𝛾2. These are expected to be significant and negative. However, to establish a causal relationship between the instrument and any outcome variables, two conditions should be met (Stock & Watson, 2013):

(20)

16

 The first is the Relevance Condition that 𝑦1 ≠ 0, or that the coefficient of interest is significantly different from zero, after controlling for all relevant firm characteristics.

 The second is an Exclusion Restriction, under which there should exist no covariance between the instrument and the error term: Cov(𝐹𝐴𝑆 𝑏𝑒𝑓&𝑎𝑓𝑡𝑒𝑟𝑖,𝑡 , 𝜀𝑡) = 0. This would mean that the only thing affecting the outcome variables are the randomized fiscal year end quarters and there are no unobservable other factors driving the results.

The first condition appears easy to discern (as visible from the results), yet the second is more difficult to straightforwardly satisfy. To attempt and prove it is complied with, the paper introduces several placebo and robustness tests, presented later in the paper.

It is also to note that the information contained in the coefficient of interest in any of the regressions presented in this paper shows the Local Average Treatment Effect of complying with FAS123R. Since each firm had different level of stock option expense, the coefficient 𝑦1 does not present a homogenous treatment effect, but only the average of the subsample of firms complying with FAS123R as it was legally prescribed. A description of the variables used in equations (1) and (2) follows below.

For the quarter of the immediate compliance with the standard, company specific controls are captured in 𝑋𝑖,𝑡. Lagged values of some of the controls are added in different specifications and in (1) and (2) are represented by 𝑋𝑖,𝑡−1 . In some regression specifications, there is a variable indicated as [t+1], the quarter post the acceptance of the accounting change & simultaneous with the hypothesized reductions of profitability. The [t+1] control is the Market Premium, other variables of the same timing are avoided to prevent structural correlations. Values for all regressors and outcome variables are taken quarterly. In addition, Industry fixed effects are captured in 𝜇𝑡. Exception make the tests of the hypothesis for the technological industry, where the regressions are ran conditionally for technological firms. The latter are indicated with a dummy variable equal to 1, if the respective firm is technological and 0 otherwise. Year fixed effects are indicated with λ𝑡, quarter fixed effects with 𝜈𝑡, but some

(21)

17 regression specifications are shown without them, due to detected collinearity. The residual is captured in the error term 𝜀𝑡.

The methodology for testing the hypothesis that shareholder returns are impaired due to lowered performance metrics further proceeds with building the second stage of an instrumental variable regression. The 2SLS has Total Shareholder Returns as the outcome variable, and the main regressor is the Ebitda Level, instrumented with FAS123R before and after in equation (3). Equation (4) has similar construct, except that it employs the fitted values of Net Income level as main regressor. The estimation of either is expected to show that the profitability levels do indeed translate in a negative effect on Total Shareholder returns. The regressions are implemented with both earnings measures, resting on the idea that Ebitda might provide a clearer picture since it is directly linked with the operations of the business. The empirical specifications look as follows:

(2SLS) (3)𝑇𝑜𝑡𝑎𝑙 𝑆ℎ𝑎𝑟𝑒ℎ𝑜𝑙𝑑𝑒𝑟 𝑅𝑒𝑡𝑢𝑟𝑛𝑠 _𝑖,𝑡+1= 𝑔₁∗ 𝐸𝑏𝑖𝑡𝑑𝑎 𝐿𝑒𝑣𝑒𝑙̂ _𝑖,𝑡+1+ 𝑔₂∗ 𝑋_𝑖,𝑡+1+ 𝑔₃∗ 𝑋_𝑖,𝑡+ 𝑔₄∗ 𝑋_{𝑖,𝑡−1}+ 𝜇_𝑡+ λ_𝑡+ 𝜈_𝑡+ 𝜀_𝑡 (2SLS) (4)𝑇𝑜𝑡𝑎𝑙 𝑆ℎ𝑎𝑟𝑒ℎ𝑜𝑙𝑑𝑒𝑟 𝑅𝑒𝑡𝑢𝑟𝑛𝑠 _𝑖,𝑡+1= 𝑔₂∗ 𝑁𝑒𝑡 𝐼𝑛𝑐𝑜𝑚𝑒 𝐿𝑒𝑣𝑒𝑙̂ _𝑖,𝑡+1+ 𝑔₂∗ 𝑋_𝑖,𝑡+1+ 𝑔₃∗ 𝑋_𝑖,𝑡 + 𝑔₄∗ 𝑋_{𝑖,𝑡−1}+ 𝜇_𝑡+ λ_𝑡+ 𝜈_𝑡+ 𝜀_𝑡 𝐸𝑏𝑖𝑡𝑑𝑎 𝐿𝑒𝑣𝑒𝑙̂ _𝑖,𝑡+1 is the fitted value of the performance metric, instrumented with FAS Before

and After, as from the 1st_{stage regression. Thus the coefficient of interest are 𝑔}

1 𝑎𝑛𝑑 𝑔2, expecting them to be significant and negative. The remaining firm specific controls are the same as in the first stage regressions. So are the fixed effects. The methodology for developing the regressions testing the remaining hypotheses is rather similar, yet details are provided below.

(22)

18 The following hypothesis tests how the analysts’ EPS Forecast Errors were affected by the acceptance of the new accounting change. A 2SLS showing the effect of the instrumented Forecast Errors on the Stock Returns is also shown. However, the cleaner and stronger effect of FAS123R is exemplified in running the previous set of regressions. Thus specification (6) only provides the partial contribution of the forecast error to the reduction of stock returns and not the ubiquitous effect of the standard’s acceptance.

(1st_Stage) (5) 𝐹𝑜𝑟𝑒𝑐𝑎𝑠𝑡 𝑒𝑟𝑟𝑜𝑟𝑖,𝑡=0= 𝜋₁∗ 𝐹𝐴𝑆 𝑏𝑒𝑓& 𝑎𝑓𝑡𝑒𝑟_𝑡=0+ 𝜋₂∗ 𝑋_𝑖,𝑡+1+ 𝜋₃∗ 𝑋_𝑖,𝑡=0+ 𝜋₄∗ 𝑋_{𝑖,𝑡−1}+ 𝜇_𝑡+ λ_𝑡+ 𝜈_𝑡+ 𝜀_𝑡 (2SLS) (6)𝑇𝑜𝑡𝑎𝑙 𝑆ℎ𝑎𝑟𝑒ℎ𝑜𝑙𝑑𝑒𝑟 𝑅𝑒𝑡𝑢𝑟𝑛𝑠 𝑖,𝑡+1= 𝑔1∗ 𝐹𝑜𝑟𝑒𝑐𝑎𝑠𝑡 𝑒𝑟𝑟𝑜𝑟̂ 𝑖,𝑡=0+ 𝑔2∗ 𝑋𝑖,𝑡+1 + 𝑔3∗ 𝑋𝑖,𝑡+ 𝑔4∗ 𝑋𝑖,𝑡−1 + 𝜇𝑡+ λ𝑡+ 𝜈𝑡+ 𝜀𝑡

In specification (5), which is the first stage of the regression analysis, the dependent variable is the Forecast Error in time [t=0] relative for Net Income in time [t+1]. The variable represents the difference between the forecasted and the actual EPS value per company per quarter. If the coefficient of this first regressor is than positive, than there in an observed increase in the forecast error attributable to the stock option expensing. The description of the controls is analogous to the one from the previous specifications.

Equation (6) represent the hypothesis that TSRs reacted to the increase in Forecast Error. Here, Total Shareholder Returns are regressed on its fitted values. This equation is further empirically estimated, however, for two different values of Total Shareholder Returns: firstly for returns adjusted for dividends, secondly, for returns adjusted for dividends and share repurchases. These outcome variables are both indexed with time [t+1], so that these are the values from the quarter following the adoption of FAS123R. The coefficient of interest here is the

(23)

19 one of the fitted value of Forecast error. If it is negative, than the excessive confusion that prevailed on the market among the analysts also translated into a loss of shareholder value. With the segregation of the outcome variables depending on share repurchases, the analysis establishes if the supposedly increased use of share repurchases actually reduces the hypothesized adverse effect of the increased forecast error. For comparative purposes a table is included juxtaposing the results of the 2SLS regressions of TSRs firstly on the affected Ebitda level from specification (2) and secondly, the affected analysts’ forecasts. Results are discussed in the following section of the paper.

3.6.2. Secondary hypothesis

For the hypothesis aiming to establish whether the mandatory stock option expensing resulted in any significant changes of analysts’ buy-sell recommendations, the dependent variable takes values from 1 to 5, as described above:

(1st Stage) (7) 𝑅𝑒𝑐𝑜𝑚𝑚𝑒𝑛𝑑𝑎𝑡𝑖𝑜𝑛_𝑖,𝑡+1 = 𝑏₁∗ 𝐹𝐴𝑆 𝑏𝑒𝑓𝑜𝑟𝑒 𝑎𝑛𝑑 𝑎𝑓𝑒𝑟_𝑡=0+ 𝑏₂∗ 𝑋_𝑖,𝑡+1+ 𝑏₃∗ 𝑋_𝑖,𝑡=0+ 𝑏₄∗ 𝑋_{𝑖,𝑡−1}+ 𝜇_𝑡+ λ_𝑡+ 𝜈_𝑡+ 𝜀_𝑡 (2SLS) (8)𝑇𝑜𝑡𝑎𝑙 𝑆ℎ𝑎𝑟𝑒ℎ𝑜𝑙𝑑𝑒𝑟 𝑅𝑒𝑡𝑢𝑟𝑛𝑠 _𝑖,𝑡+1= 𝑐₁∗ 𝑅𝑒𝑐𝑜𝑚𝑚𝑒𝑛𝑑𝑎𝑡𝑖𝑜𝑛̂ _𝑖,𝑡+1+

𝑔

₂

∗ 𝑋

_𝑖,𝑡+1+ 𝑔₃∗ 𝑋_𝑖,𝑡+ 𝑔₄∗ 𝑋_{𝑖,𝑡−1}+ 𝜇_𝑡+ λ_𝑡+

𝜈

_𝑡+ 𝜀_𝑡 The regression specifications further entail the following variables. The outcome variable in (7) is the Consensus Recommendation value for time [t=1], which is the exact quarter of interest. The main coefficient of interest if of FAS123R before and after [t=0]. A negative value of it would indicate that the consequent recommendations were downwardly revised, as hypothesized.

(24)

20 Considering that recommendations might be changing for reasons different than the standard, leads to the following brief discussion: if a poorly governed firm, because of the poor management, engages in e.g. myopic practices (Jochem, Ladika & Soutner, 2015), than its corresponding recommendation might be timely revised to allow for that. Or if the company already has very low profits, the generation of such an expense as the stock options might results in significant negative returns; which might in turn translate in its recommendation downgrade. Such unobservable factors admittedly have the potential to bias an empirical discovery, but are hereby minimized, due to the identification of the exact timing of FAS123R taking effect and employing it as the instrumental variable in each regression. In addition, the regressions empirically control for a number of confounding variables & the results are reasonable according to theory and consistent across different papers. The latter words additionally vouch for the internal validity of the established empirical models, besides the unique setting of the uncorrelated-with-the-outcomes FAS123R acceptance.

Although these equations show a 2SLS analysing the effect of changes in recommendations on firm value, literature suggests that no such effect is expected. Therefore the hereby presented hypothesis is the same: a two stage least squares regression of stock returns on the recommendations instrumented with FAS before and after should provide no significant results. Indeed running the regression finds none, especially when controlling for time fixed effects.

3.7. Data & Summary statistics

Data is collected from the cross-section of Compustat and IBES, initially starting with the entire databases for North America, using quarterly data for all dependent variables, all main regressors of interest and control variables. The two databases are accessed through the platform of Wharton Research Data Services and are widely used in financial research. Compustat provides extensive datasets covering firm fundamentals, whilst IBES is the source of information covering analysts’ forecasts, the corresponding actual values and buy-sell recommendations. The entire North America dataset of Compustat consists of 22 196 companies for the years

(25)

2000-21 2012 which fall in the scope of the present research. The IBES US file originally contains half of the firms on Compustat. After excluding firms which changed fiscal year ends shortly prior to FAS123R & those who accelerated option vesting before the mandatory compliance, the number of unique firms in the present sample equals 70953_.

Table 1, Panel A shows the most relevant summary statistics of the control variable. The values are for the years around FAS123R acceptance, namely 2005-2007. Exception are the values for the Analysts’ Recommendations which are given in a larger time span 2003- 2008, for the sake of increasing the number of observation in each regression (since the number of available analyst recommendations in the IBES database proved to be lower than other data availability). The table shows that the number of observations available per control variable varies between 21 000 and 27 000 observations. The average firm in the sample is characterised by stock return of 2%, negative Ebitda-to-Assets of -0.01 & Net Income-to-Assets of -0.02. The average Book-to-Market ratio is 1.88. The stock price volatility is on average 53.42%.

Panel B focuses on the Analyst variables: the average EPS actual value is 0.22, while the analysts would on average forecast it at 0.15. There are no “Strong buy” recommendations in the sample and the average mean Recommendation is 2.36, while the consensus median is 2.4.

Panel C shows the distribution of fiscal year endings which permit the establishment of the exogenous dummy variable FAS123R. As in e.g. Jochem, Ladika & Soutner (2015), in the present paper there is also sufficient cross-sectional variation in the timing of the companies’ fiscal year filings. Most of the firms submit their final reports in December, 73%. As also visible from the table, approximately 11% of the firms have fiscal year end before the month of May. During Quarter 1 of each year, 8,5% of the firms file their annual reports, in Quarter 2, these are 8.7%, in quarter 3 – 7.5%. 75% of the companies do so in Quarter 4.

(26)

22 Panel D shows the unconditional means, standard deviations & quantile statistics for all the firms in the period 2000-2005 before the introduction of the accounting change. These values are relative for the analysis of the regressions presented in the upcoming tables and would be referred to on the following pages. Juxtaposing for example the Stock Returns in this period with the Stock returns in 2005-2007, shows a drop of the mean in the consecutive years. If before the mean Return would have been 4%, than it is halved in the scrutinised period down to 2% for the average firm. The Net Income-to-Assets of the average firm also diminishes from negative 0.02 to -0.03. Another observation is that the mean Forecast Error based on Median Consensus EPS estimates increases from 0.07 to 0.16.

4. Empirical Results and Analysis

4.1. Key results from main hypotheses

The analysis of the spill over effects of the mandatory compliance with FAS123R begins with testing whether its adoption had any effect on the companies’ profitability. In other words, the above mentioned changes in the characteristics of the average firm are scrutinized to establish causality & magnitude of the compliance with FAS123R. The main regressor are the random fiscal year ends of firms, translated to compliance-with-the-standard quarters, which are marked with 1 in all quarters of & post compliance, and 0 prior to that. This is the FAS123R before

and after variable in the paper. The first stage regressions establish if it affected Ebitda and Net

Income levels & the analysis further proceeds to a 2SLS of Stock Returns on profitability.

4.1.1. Fiscal year ends, Ebitda & Net Income response to FAS123R

Table 2 shows the effect of FAS123R on the Ebitda level in specifications (1) to (3), and on Net Income level, in (4) through (6). Each specification shows negative association between the main regressor FAS123R before and after and the outcome variables, confirming the hypothesized outcome and in agreement with the changes of the means of firm characteristics. Specification (1) and (4) are only regressing the dependent variables on the FAS123R dummy,

(27)

23 while the rest of the regressions introduce controls. The result is barely changed after controlling for firm size, indebtedness, previous stock performance and volatility, employee stock option grant values. Specification (2) and (5) introduce the full set of control variables, controlling for industry fixed effects and year fixed effects. All columns focus on the period around the acceptance of FAS123R, i.e. calendar years 2005-2007. Thus the focal point is the mandation of stock option expensing, when any effect of the accounting change should be mostly expressed. The sample of firms is the cross-section of IBES and Compustat, and is further limited by the availability of quarterly data on controls.

Since neither the outcome variable nor the main right hand side variable is in logarithm, and the regressor is in fact a 0/1 dummy variable, the interpretation of the coefficient is rather straightforward (Stock & Watson, 2010, pp.250 -260). The value is in the units of the outcome variable, and allows for a direct inference of its effect on the unconditional mean value of e.g. Net Income/Assets from before the acceptance of the standard. Thus, since the unconditional mean from the time span 2000-2005 was -0.02, it would drop to a -0.03, as the coefficient in specification (5) is -0.01. This is a 50% drop in the unconditional average Net Income level of a firm attributable to FAS123R (and is the Local Average Treatment Effect, only for the group of compliers).

If the time span is reduced and the comparison is narrowed to quarters immediately prior compliance, after 2005 and before 2007,& for the observations only falling within the regression, the effect is lesser. When related to the constant of the regression, returning the within-sample average, Net Income drops 15% (calculated as the marginal effect -0.011/-0.071). Similarly calculated, the Ebitda drops 17%, as inferred from specification (2). Both figures show significance at the 1% level. These results, however, are not controlling for time fixed effects. When introducing year & quarter fixed effects, the respective coefficients drop to -0.006 in specifications (3) and (6), almost twice as low. Yet, these retain significance at the 10% level. Thus the economic magnitude of the detected effect of FAS123R is put forward namely in these

(28)

24 specifications (3) and (6): i.e. the mean Ebitda level decreases with 11% and the Net income level with 9%. As hypothesized, the effect is clearer for Ebitda, as it is more affected by FAS123R.

To further clarify the results, the behavior of the F-statistic should as well be discussed. It is the Kleibergen-Paap (2006) F-test for significance & weakness of the instrument (or the main regressor in the present case). As Stock, Wright & Yogo, (2002, pp. 518 -529) discuss, the scale of this F-statistic is the same as for the joint-significance of the regressors: the threshold is 10 for significance at the 1%. From table 2 it can be read that in specifications (1), (2), (4), (5), the F-statistic of the instrument exceeds the threshold, thus suggesting the significance of the instrument.

However, the inclusion of time fixed effects reduces its strength, bringing down the F-statistic below 10, supposing that the dummy variable for the acceptance of FAS123R likely does not explain a significant part of the variation in the outcome variables. This could already be read as violation of the Relevance condition for the instrument & has the potential to limit the inference possibilities for the actual magnitude of its effect. Therefore, the p-values (of the test) can be considered for further clarification. What they show is the following: although the F-statistic if below 10, the p-value (p-value>F) is 0.0161 in e.g. specification (3). This value shows that the Null hypothesis (that the instrument is weak) can be rejected at the 5% level. What is more, Chouhary, Rajgopal, & Venkatachalam, (2009) and Balsam, Reitenga and Yin (2006), who also delve into the spill-over effects of FAS123R, also rely on these p-values (where they use a similar instrument).

Contrastingly to them, Ladika & Soutner (2014) have F-statistics of their instrument (which is similar to the present) well above the threshold even when they consider year fixed effects. Their study is conducted on yearly data. Consequently, an experiment was done here aiming to establish the behavior of the F-statistic had the present survey also relied on annual data. In the same regression specifications, if ran with all variables annual, the F-statistic was significantly higher. Since these regressions were not done extensively and are not reported, they remain subject to a different study and further investigation. However, the finding that the

(29)

25 quarterly analysis is affected more strongly by time fixed effects is thus clearly delineated. A potential reason can be sought in the high correlation coefficients between the FAS123R before

and after dummy and the year and quarter fixed effects (as visible from the correlation table 9).

This high correlation is also explainable with the fact that the mandatory compliance with the standard highly depended on the timing of firms’ fiscal year ends. Thus compliance occurred staggered in different quarters. Hence, their inclusion can be blurring the interpretation. However, relying on the brief discussion from above, putting forward the significant p-values of the F-test is going to be the central argument as to why the coefficients in specifications with time fixed effects are to be henceforth interpreted as significant, depending on the p-value of the F-statistic.

Another issue is the joint significant of the regressors, which is satisfied throughout all specifications in all tables. This F-statistic is not reported in favor of the more important one - of the instrument.

4.1.2. Effect of FAS123R through Ebitda & Net Income, on Stock Returns: Second

Stage

The analysis further proceeds to establishing the effect of FAS123R, through the affected Ebitda & Net Income on the Stock returns. Specifications (1) through (4) in Table 3 have the Ebitda Level as the main regressor, while (5) through (8) have the Net Income level as such. The period scrutinized in each is 2005-2007 & the number of observations depends on data availability for all controls. The baseline sample consists of all firms in the cross-section of Compustat & IBES, for firms which were founded in the US. All coefficients discussed are significant, mainly at the 5% level. Exception is specification (8), which is only significant at the 10%.

In specification (2), the 2SLS finds that a one standard deviation downward change of Ebitda Level led to a (0.09*11.788) = 1.06092% absolute drop in stock returns adjusted for dividends, which is a 26.5% decrease relative to the unconditional mean of 4%. In the same time,

(30)

26 returns adjusted for dividends and repurchases declined by 1.0612%, translating from specification (4).

Similar calculations for the Net income level translate into a (0.12*10.888) 1.3056% absolute decrease of the Stock returns from the unconditional mean of 4% (reading from specification 6). These are a 32,64% relative drop of Stock Returns for the average firm in the sample. The findings presented above concern Stock Returns adjusted for dividends and also prove the initially stated hypothesis that FAS123R would exert an adverse effect on the investment returns. Scrutinizing the effect of FAS123R, through the Net Income, on Stock Returns adjusted for share repurchases, results in s 32.68% in the returns (specification 8).

The second hypothesis in this relation concernes the potential of stock repurchases to moderate the adverse effect of FAS123R, as it was suggested by Goldman & Kohlback (2009). Contrastingly to what was expected, however, specifications (4) and (8), prove the opposite. Stock returns adjusted for dividends and repurchases were more affected by the standard. For example, if Stock returns adjusted for dividends declined by 1, 0609% (specification 2) due to the drop in Ebitda level, the Stock returns taking into consideration the buybacks dropped more: by 1.0612% (specification 4). This, however, is established to highly depend on the exact regression specification. The trend is reversed if additional regressors are added. This is further discussed later in the paper & a robustness check is provided.

All the specifications referred to control for industry and time fixed effects. The addition of the latter, as discussed in the previous section, reduces the F-statistic of the instrument, but the p-value of the F-statistic remains significant at the 5% level for the Ebitda regressions & at 10% for the regressions employing Net Income Level as main variable of interest. Table 3 juxtaposes each regression with and without time fixed effects, so that the variation of the coefficients can be tracked. This pertains to the fact that Operating Income is likely the better measure of a firms’ performance & provides a clearer view on the scope of FAS123R effects.

(31)

27

4.1.3. Before and after Effect of FAS123R on Analysts Forecast Error: First Stage

Table 4 analyses the effect of the accounting change on the Analysts’ ability to forecast Earnings per share. Specifications (1) though (4) concern all firms for which there is available data; specifications (5) to (8) focus on the technological industry in particular (since heavily reliant on option based remuneration). The forecast error which is the outcome variable in each of the regressions is computed in two manners. Firstly, the Median consensus EPS forecasts are used, to measure their difference from the actual value. This is named Forecast Error (by median

estimate) and is the outcome variable in specifications (1), (2), (4) and (5). Secondly, the forecast

error is computed on the basis of the Mean consensus estimates, and is employed in the rest of the specifications (Forecast Error by mean estimate). This segregation was found necessary since in the literature the Median value is the favoured one, but the reactions of either to FAS123R has not been scrutinized so far. Therefore, the analysis clarifies which is the measure less susceptible to influence.

Since again neither the outcome nor the main independent variable is in logarithms and the main regressor is a dummy variable, the interpretation of the coefficients is rather simple. There are two venues for relating the forecast error attributable to the FAS123R before and after affect (Stock & Watson, 2010). One: it can be taken relative to the unconditional mean Forecast error (by median), which is 0.07, or by median, which is 0.08 (from the period 2000-2005). Another: it can be taken relative to the constant term in the regressions, which represent the Forecast errors within the observations under each regression specification & is from the period 2005-2007. The latter method is adopted, since is analyses a more concrete time span concentrated around the adoption ofFAS123R.

A delineation between regressions with and without time fixed effects is also made, each second regression having year fixed effects. Due to collinearity detected between quarter fixed effects and the market premium, quarter fixed effects are omitted in favour of the market premium as a significant regressor. Year Fixed effects are retained. The F-statistic in each regression is higher than the threshold of 10.

(32)

28 The first formal discovery is that Forecast Error based on median estimates was more susceptible to the influence of FAS123R, i.e. forecast error (by median) increased with 0.036 (specification 2), compared to 0.035 (specification 4) for all the firms in the sample. Exactly the opposite is observed for technological firms. Not only was the magnitude of the error higher, but the mean estimates were less on par with reality: 0.045 (specification 8) compared to 0.044 (specification 5) by median. In relative terms, these values have the following meaning:

Considering the more erroneous median estimates for all firms, the average error before the adoption of FAS123R was 0.067(specification 2). This sharply increased by 0.036, or by 53%. For the technological firms, the error increased by 52%, but the threshold error level was higher: 0.087 forecast error (specification 8). Considering that the mean actual EPS value for the average firm was 0.22 in the same period, this means that analysts were overoptimistic enough to overstate EPS forecast to 0.32. In other words, they thought that the Earnings Per share of the average firm was 46% higher. For the technological firms, the overstatement was higher. The actual EPS value of the average Tech firm was 0.24. The forecasted value after the adoption of FAS123R was 0.37, which was a 55% overstatement.

These findings vouch for the analysts’ distraction and for the fact that they could not envision how would firms respond to the mandatory stock option expensing. FAS123R affected their estimates as well as the companies’ profitability.

4.1.4. Effect of FAS123R through Forecast Error on Stock Returns: Second Stage

Table 5 shows the effect the increased Forecast Errors had on Stock Returns. The time span in each regression is 2005-2007 & specification (1) through (4) focus on all firms in the sample for which data is present & specifications (5) through (8) analyse the situation of technological firms in particular. Two measures of Stock Returns are again introduced: in specifications (1) and (2), (5) and (6), the Stock Returns are considered only as adjusted for dividends, while in the rest of the regressions, outcome variables are the Returns adjusted for dividends and Stock Repurchases. The construct is rather similar to the one discussed in section

(33)

29 4.2., table 3. Each second regression is shown with time fixed effects so that their effect on the coefficients is clearly visible. These are alto the regression specifications discussed in detail on the following lines. The F-statistic in each is higher than the threshold of 10.

Specification (2) shows that a one standard deviation increase in the forecast error (0.45*-1.597) resulted in a 0.72% decrease in the stock returns adjusted for dividends. This, relative to the unconditional mean of 4% amounts to an 18% drop in stock returns for the average firm in the sample. Stock returns adjusted for dividends and repurchases (specification 4) declined by 17.8%, i.e. they were less impacted by the accounting change.

Taking under scrutiny the results for the technological industry, the following findings come interesting. A one standard deviation increase of the Forecast error (by median) (0.39* - 1.166) amounts to a 0.454% (column 6) absolute decrease in Stock Returns adjusted for dividends & slightly more when adjusting for dividends and stock repurchases: 0.459% absolute decrease (column 8). Since the mean returns for the tech firms were approximately 1% in the period 2000-2005, the above mentioned declines in percentages are also the relative changes attributable to the adoption of FAS123R.

The overview of the results is thus in unison with theory: since the technological industry is more intensive in remunerating its employees with stock options, and thus more affected by the FAS123R, the percentage changes exhibited by these firms is higher, compared to the average firm in the sample. This is, however, only true from the analysts perspective. From the perspective of Company Profitability, the opposite was discovered. The results add to the volume of literature tracking Analysts influence on the stock performance & are in support of them. This fact comes natural, knowing that investors rely on their estimates and are prone to unwind their shareholding when a firm EPS estimates is downgraded. Investors would perceive this as a potential signal of their stocks losing value and act to prevent potential losses.