Equity Market Timing in the U.K.
A thesis submitted in partial fulfillment of the
requirements for the degree of
Master of Science in Business Administration
2
Equity Market Timing in the U.K.
ABSTRACT
This paper researches the prevalence of equity market timing in the U.K., studying a sample of 250 British firms that were active throughout the period 1985-2009. Regressions of equity capital flows on proxies for market timing opportunities cannot support a hypothesis of managers attempting to time the equity market. Internal rate of return-based measures of market timing show large variations across the sample, but also indicate that the average firm was not successfully timing the market. Nevertheless, I find that equity capital flows help predict future stock returns in the cross-section, in ways that are consistent with the market timing hypothesis. Combining the results, I conclude that market timing may not play the significant role in equity markets that is claimed in much of the earlier literature. Earlier studies on long-run returns following equity issuance may have been wrong in attributing their findings to market timing.
Keywords: equity market timing, market efficiency, capital structure, dollar-weighted returns, abnormal returns, seasoned equity offerings, share repurchases
3
Modigliani and Miller‟s (1958) basic theorem of capital structure states that in an efficient and frictionless capital market, the value of a firm is not affected by the way in which a firm is financed. But when we relax the improbable assumption that market prices always fully reflect all publicly and privately available information, financial policy does become relevant for market participants in obvious ways (see e.g. Baker, Ruback and Wurgler (2007), Shefrin (2001)). Corporate managers with insider information may, for instance, be able to adjust the capital structure of their firm in order to benefit ongoing, or „buy-and-hold‟ shareholders at the expense of entering and exiting ones, that is, they may be able to engage in what we refer to as „equity market timing‟ practices. To date, there has been much disagreement in the financial community and amongst academics as to whether corporate managers are really taking advantage of private information in such ways. My objective in this paper is to bring more clarity to the field, by studying the topic of equity market timing from several angles.
As Myers and Majluf (1984) and DeBondt and Thaler (1987) explain, asymmetric information about the value of the firm can induce managers to engage in market timing practices. Managers may, for example, issue new shares when they feel their existing shares are overvalued by the market compared to their actual, „true‟, value. If the stock price does not move to its fundamental value in response to the announcement of a new issue, this means that the stock stays overvalued and existing shareholders benefit at the expense of new investors, who thus paid too much for their shares. In a similar vein, managers may repurchase shares when they feel their firm‟s shares are worth more than other market participants are willing to offer. If the share price does not increase to fundamental value in immediate response to the repurchase announcement, remaining shareholders benefit at the expense of those who sell their shares to the firm. Assuming that the stock price eventually converges to its fundamental value, successful market timing implies that issuing (repurchasing) firms will have below (above) average long-run returns.
In fact, starting with Ritter‟s (1991) seminal study, several papers have documented abnormal returns following stock issues and repurchases (see e.g. Baker, Ruback and Wurgler (2007), Pontiff and Woodgate (2008), DeAngelo, DeAngelo and Stulz (2010), among others); findings that are supportive of the hypothesis that successful market timing takes place. Additional lines of evidence for market timing are provided by management surveys (Graham and Harvey (2001), Brav et al. (2003)) and studies of insider trades (Jenter (2005)).
4
may be lower than what they are commonly estimated to be – by looking at weighted-average simple stock returns. The reasoning behind this is that the actual returns earned by investors depend not only the buy-and-hold returns of the equities invested in, but also on the timing and size of cash flows into and out of these securities. Dichev (2007) puts this into perspective and applies an internal rate of return measure, taking account of aggregate equity inflows and outflows, to U.S. data. His findings indicate that investors may fall short by about 1.5% yearly due to corporate managers engaging in market timing.
If such findings as Dichev‟s (2007) are reliable and if successful market timing is indeed taking place, as the body of evidence on long-run abnormal returns following equity distributions indicates, this has important implications for all equity market participants and for the financial community at large. It would mean that companies‟ costs of capital are lower, which in turn implies that the market equity risk premium is lower, than hitherto estimated (see Baker, Ruback and Wurgler (2007)). It also has practical investment implications, indicating that contrarian, or at least, passive, investment strategies are likely to do well.
But despite the body of evidence indicating that market timing plays a significant role in the equity market, the issue of whether firm financial managers are successfully timing the market is still a subject of considerable debate amongst academics and finance professionals. For several plausible reasons, many researchers continue to question the validity of the market timing model and tests thereof. Some authors highlight the joint hypothesis problem: in order to show that returns are „abnormal‟, they have to be compared against a model that describes the required rate of return. As long as we are not sure which model correctly stipulates the returns required by the market, we cannot determine with certainty whether and when returns are abnormal (see Fama (1991), Eckbo, Masulis and Norli (2000), Brav, Geczy and Gompers (2000)). Other authors point out that exaggerated expressions of abnormal performance are obtained when abnormal returns are measured in „event time‟ instead of calendar time (Schultz (2003), Butler, Grullon and Weston (2006)). Keswani and Stolin (2008) place serious doubts on Dichev‟s (2007) findings in particular, arguing that they are not robust to changes in time periods and markets.
5
In light of the fact that, as mentioned above, equity market timing is still a highly debated and controversial issue, the central question of this study is:
How prevalent is the practice of equity market timing in the U.K.? To answer this question I consider in turn the following three sub questions:
1. Can we observe U.K. firms attempting to time the equity market? 2. a) To what extent are U.K. firms successfully timing the equity market?
b) What distinguishes those firms that are (more) successful at timing the equity market?
3. Do patterns of equity capital flows in the U.K. provide information to predict future stock returns?
The answers to these three sub questions together help to determine whether U.K. corporate managers are purposefully and/or successfully timing the equity market. The data used for this study consists of monthly financials of 250 large British firms retrieved from Thompson Datastream, covering the period from January 1985 through December 2009.
6
fourth, this is the first research that investigates differences in timing abilities in the cross-section of firms. I specifically look at whether market timing successes differ with firm size, book-to-market, trading volume and operating industry.
The remainder of this paper is organized as follows. Section I presents an overview of the main literature considering equity market timing. Section II describes the collection procedure of the data used throughout the rest of the paper. Sections III, IV and V address in turn the three sub questions presented above, and section VI synthesizes the findings and concludes.
I. Literature Review
A. Empirical Evidence Regarding Equity Market Timing
A vast body of literature exists with respect to equity market timing. Although a considerable amount of findings suggests that managers are regularly – and successfully – engaging in market timing practices, several academics claim that these findings are due to flaws in research designs. This subsection sets out the most important findings and theoretical contributions regarding the topic of market timing to date. For clarity purposes, the studies referred to are presented in chronological order in appendix A.I., showing the stock markets studied, the sample size and the research methodologies along with the main findings.
A.1. Managerial Ability to Time the Equity Market
Several lines of research suggest perceived stock misvaluation to be a motive for equity issues and repurchases. Most straightforwardly, in a survey of 392 CFOs by Graham and Harvey (2001), two-thirds convey that “the amount by which the firm‟s stock deviates from the perceived true value is an important consideration when issuing equity” (pp. 216). Similarly, of 384 financial executives surveyed by Brav et al. (2005), 87% agrees that “we repurchase when our stock is a good investment, relative to its true value” (pp. 493).
7
those of size matched control firms. Levis (1995) finds similar patterns in U.K. equity data. On the other hand, share repurchase programs – which can be viewed as „negative stock offerings‟– tend to be followed by positive abnormal returns. In their 1980s sample of 1,200 open market repurchases, Ikenberry, Lakonishok and Vermaelen (1995) find repurchasing firms to significantly outperform matched firms by 12% over the four years following a repurchase announcement. Consistent evidence is found by Chan, Ikenberry and Lee (2007), who examine 5,508 U.S. repurchase programs between 1980 and 1996; similar results are shown for the equity market of Japan in a recent paper by Ishikawa and Takahashi (2011).
Furthermore, it appears that the timing ability of managers is not restricted to pure equity-versus-cash transactions; it extends to mergers and acquisitions that take place in exchange for stock. Loughran and Vijh (1997) investigate a sample of 947 U.S. acquisitions that took place between 1970 and 1990 and find that acquiring firms that completed stock mergers experienced negative long-run excess returns of 25%. In contrast, they find that firms that completed cash tender offers outperformed matching firms by as much as 61.7% over the same time horizon.
Yet other academics suggest that when abnormal returns are properly measured, evidence for long-run abnormal returns subsequent to equity capital flows disappears (i.a. Eckbo, Masulis and Norli (2000), Brav, Geczy and Gompers (2000)). Most significantly, Eckbo et al. (2000) argue that the much applied matched-firm technique fails to properly control for risk. Under the matching technique, every (in this case, issuing or repurchasing) firm in a sample is matched to a (non- issuing or repurchasing) peer on the basis of market value or book-to-market ratio, and abnormal returns are then calculated as the difference between the returns of the sample firms and those of their peers. Eckbo et al. (2000) point to the fact that equity issues and repurchases change firm leverage and influence stock liquidity, which may in turn lead to changes in equity beta and risk. A firm that issues new stock, for instance, will have a lower leverage and a higher stock liquidity than an otherwise identical peer firm that does not issue. The issuing firm will as a result be less risky than the peer, and its investors will require a lower return. Eckbo et al. (2000) claim that most research findings regarding managerial market timing ability may be invalid, because researchers have largely failed to incorporate these effects in their testing methodologies.
8
traditional event studies treat timing of events as exogenous and consequently, ex post, the issuers seem to time the market by the finding that „the average firm going public underperforms the market in the long run‟. According to Schultz, the methods by which abnormal returns are measured can cause significant inference problems. He postulates that performance differences between issuing and non-issuing firms will disappear when returns are measured in calendar-time instead of in event-time. While event-time methods implicitly test a strategy of investing equal amounts in each offering, calendar-time methods test a strategy of investing equal amounts in offerings each month. Indeed, most research papers into equity market timing before the publication of Schultz‟s paper make use of event-time abnormal return measures (i.a. Ritter (1991), Loughran and Ritter (1995), Ikenberry, Lakonishok and Vermaelen (1995), Loughran and Vijh (1997)). In essence, Schultz‟s (2003) theory, which is dubbed the „pseudo market timing hypothesis‟, entails that managers cannot purposefully engage in market timing because they are unable to determine whether their stock is over- or undervalued at any time (i.e. their expected future abnormal return is always zero); instead they just regard a high price of their stock relative to the past stock price pattern to be a good moment to issue new stock.
Schultz‟s (2003) pseudo market timing theory quickly raised the interest of other academics. Butler, Grullon and Weston (2006) find evidence consistent with the pseudo market timing hypothesis for the U.S. market. On the other hand, Dahlquist and De Jong (2008) extend the research with more advanced econometric simulation studies, accounting for the endogeneity of the equity issuers‟ timing choice, and find only a minor bias due to Schultz‟s hypothesis. They therefore regard it unlikely that the underperformance of issuing firms can be attributed to pseudo market timing. Their findings are further supported by Ang, Gu and Hochberg (2007) and Chan, Ikenberry and Lee (2007). However, Gompers and Lerner (2003) show that results of long-run studies may be time period sensitive, since they find no evidence for abnormal performance when they apply the calendar-time method to IPOs in the earlier 1900s.
9
targeted at a firm‟s ongoing shareholders – and firm commitment offers. Long-run underperformance is found only in the firm commitment group, indicating that managers try to profit from new shareholders only.
To sum up, academics have yet to reach agreement on the topic of managerial timing ability and the controversy may never be fully resolved. But, considering all these studies collectively, the evidence points in the direction of market timing playing a nontrivial role in the equity market.
A.2. Factors Influencing Equity Distributions
Firms can have many different reasons for issuing or repurchasing equity. For example, a firm might want to change the level of leverage, or raise new capital in order to exploit investment opportunities. Autore, Bray and Peterson (2009) investigate the relation between the intended use of proceeds as stated in proxy statements and the long-run performance of SEOs. They find long-run abnormal returns only for firms that state recapitalization or general corporate purchases as the reason for raising new equity. On the other hand, they find that those firms that succeed in signaling that their reason for issuance is for specific investment purposes, do not significantly underperform in the long-run.
10
market timing opportunities – which they measure with market-to-book ratio and past stock returns – have a significant positive influence on the probability of a firm to conduct an SEO.
B. Equity Market Timing and Dollar-Weighted Returns
So far, the empirical studies indicate that market timing plays a significant role in equity distribution decisions. However, the effect that the practice of market timing may have on firm performance remains to be quantified. An estimate for the reduction in the cost of equity due to market timing may be found with the „dollar-weighting‟ of returns methodology employed by Dichev (2007)1. Dichev looks at the market timing story from the investors‟ viewpoint. He shows that the actual returns earned by stock investors are not only determined by the returns on the underlying investment, but also by the timing and amount of capital flows.
The idea behind the concept of dollar-weighted returns can be illustrated with a simple example. Consider an investor who buys 100 shares in a firm‟s IPO at €15. One year later the stock price has risen to €25 and the firm conducts a seasoned equity offering, in which the investor decides to buy another 100 shares. Then the stock price starts dropping again, till it reaches €15 at the end of the second year. Now the geometric average return (that is, the „buy-and-hold return‟, or BHR) for the stock over the two years is zero since the stock price is still the same, but it is clear that the investor has suffered a loss since he poured in extra money at a high price.
The common approach towards analyzing historical returns is to compute geometric average returns and adjust these for risk, but such measures ignore the effects of the timing of investor cash flows. Dollar-weighted returns (DWRs) rely on the concept of internal rate of return and place greater weights on time periods in which investors have more money invested. In essence, BHRs can be regarded as „time weighted‟ returns while DWRs are „time and value weighted‟. The difference between the BHR and the DWR earned over a time period, the „performance gap‟, is a potentially useful measure that gives a ballpark estimate of the effects of market timing (see also Friesen and Sapp (2007), Weinbaum (2009)).
In his research, Dichev (2007) focuses on aggregate country stock markets and shows that there exists a sizeable performance gap for investors in most major markets around the world. Furthermore, he finds strong evidence of aggregate stock returns being lower following equity contributions and higher following equity distributions. Dichev‟s results
11
contribute to the evidence that firms on average are able to lower their cost of equity by engaging in market timing.
Unfortunately, Dichev‟s (2007) research findings are fundamentally criticized a in more recent article by Keswani and Stolin (2008). They replicate part of Dichev‟s research and argue that his test results are not robust, as they appear to be very sensitive to aggregation across both time periods and markets. Even more notably, they discard all results of the international (i.e. outside of the U.S.) test results as being erroneous due to the use of national market index figures retrieved from Thomson Datastream, pointing out that the increased coverage of international stock markets in the database cause the „cash-inflow‟ measure in Dichev‟s calculations to be artificially inflated.
C. Research Hypothesis
The literature discussed so far helps to form expectations as to what may be found regarding the main research question. The quoted studies that investigate the U.K. market find similar results to those that focus on the U.S. (see Loughran, Ritter and Rydqvist (1994), Levis (1995), Henderson, Jegadeesh and Weisbach (2006)). I therefore have no a priori reasons to expect the U.K. equity market to behave differently. I tentatively hypothesize not only that British corporate managers are purposefully trying to time the equity market, but also that they are successful at this. Successful market timing means that investors as a group underreact to the information revealed by the announcement of equity capital flows, and hence that the market is not fully efficient (see Baker, Ruback and Wurgler (2007)). My expectation of successful market timing therefore also implies that knowledge of equity capital flows helps in predicting future stock returns.
II. Data Collection
12
firms having existing return index values for both dates is saved. From this list of firms I then pick those firms which are incorporated in the U.K., which can be told by inspection of the TDS variable GEOG. An inspection of the firm variable NAME shows that the list of firms includes common equities as well as preferred shares and shares with different voting rights. Since my research is focused on market timing of common equity by listed firms only, I delete those firms that have „preferred‟, „deferred‟, „pref.‟, „A‟, „B‟ or „duplicate‟ in their TDS variable NAME. I also inspect the ICB sector mnemonics of the firms and delete those that are referred to as „unquoted equities‟. Finally, I sort the firms by their market value (TDS mnemonic MV) as of January 1, 1997 and select the 250 largest firms for this date. In this way, I end up with a sample of influential UK firms, without any missing market value and/or return observations during the period from January 1985 through December 2009.
I realize that including only firms that been active throughout the entire sample period may cause a survivorship bias in my research findings, but I have several reasons for doing so. Firstly, in order to calculate an internal rate of return measure that provides a point estimate of the degree of success firms have had in timing the equity market (I do this in section IV), it is important not have any intermediate missing return and market value data. Secondly, when performing panel data regressions (as in section III and V), missing values can be ignored or they can be replaced with a proxy; but both of these remedies lead to less reliable and possibly biased results.
For all firms in the sample, I retrieve the following figures, where applicable in U.K. pounds and on a monthly basis (TDS mnemonics in parentheses) for the period from January 1, 19822 to January 1, 2010:
Total return index (RI): this gives the theoretical growth in value of a stock assuming all dividends are reinvested.
Market capitalization (MV): this is equal to the share price times the number of ordinary shares outstanding.
Book value of equity (WC03501): common shareholders‟ total investment in the company.
Trading volume (VO): the total number of shares traded for a stock.
ICB industry code (FTAG2): the industry classification benchmark sector in which a firm operates, at the lowest level of detail.
Number of ordinary shares outstanding (NOSH).
13
Factor to adjust shares outstanding (FA): a figure with which the number of ordinary shares outstanding can be adjusted for distribution events that have a limited effect on firm market value, such as splits and rights offerings.
I also collect the total return index series for the MSCI U.K. index (MSUTDKL (RI)).
III. Taking Advantage of Market Timing Opportunities
This section address the first sub question presented in the introduction. Specifically, I research whether U.K. firms have been attempting to time the equity market by investigating the relationships between equity capital flows and different proxy-measures for firm market timing opportunities. To this end, market timing opportunities are proxied by past and future abnormal stock returns and changes in firm book-to-market equity ratios. Firm market value is added as a control variable in the regression specifications. I emphasize that I do not test whether market timing opportunities drive the decision to distribute equity capital, like Loughran and Ritter (1995) posit. Instead, I aim to determine whether market timing opportunities have a significant influence on equity capital flows. I end up finding little support for the hypothesis that U.K. corporate managers have been purposefully attempting to time the equity market.
A. Regression Variables and Expected Signs
14
I consider it likely that equity distributions are also related to firm market value, I use market value as a control variable in the regressions.
An inspection of the earlier papers mentioned in the literature review reveals some relevant relationships between firm equity distributions and the proxies for mispricing mentioned above. Marsh (1982), Loughran et al. (1994), Jung et al. (1996), Hovakimian et al. (2001) and Graham and Harvey (2001) all find evidence of firms issuing equity after stock price run-ups. Pagano et al. (1998) find a negative relationship between the probability of conducting an IPO and the book-to-market ratio of peer firms, and Henderson et al. (2006) and DeAngelo et al. (2010) find a negative relationship between stock issues and future stock returns. Intuitively, changes in firm market value may proxy for stock misvaluation in a similar vein as do book-to-market ratios, that is, increasing market value may indicate overpricing and decreasing market value may indicate underpricing. On the other hand, it is likely to be relatively cheaper for larger firms to change firm leverage (Titman and Wessels (1988)), so that firm market value could be positively related to the absolute value of firm equity distributions. Also, it might be that mature firms, which tend to be larger, issue less capital since they have fewer investment opportunities (DeAngelo, DeAngelo and Stulz (2010)).
Considering the theoretical explanations as well as earlier empirical findings, I expect a measure of total equity issuance (measured as flows from investors to firms) to be positively related to past abnormal stock returns, negatively related to future abnormal returns and negatively related to firm book-to-market ratio. No specific sign is expected for the relationship between equity issuance and firm market value. The measurements of the variables that are used in this section are described below and – for convenience – also listed in appendix B.I.
A.1. Variable Measurements A.1.1. Dependent Variable Total equity capital flows
I use two different measures of equity capital flows as the dependent variable in the regressions. The first one is a measure which Pontiff and Woodgate (2008) call „composite share issuance’. This variable can be calculated as:
15
with
AdjustedSharesit = SharesOutstandingit/TotalFactorit (2)
Here ISSUEit,,it-x measures composite equity issuance for firm i between month t-x and month
t, TotalFactorit is the factor to adjust shares outstanding for firm i at the beginning of month t,
and SharesOutstandingit is the unadjusted number of shares for firm i at time t.
The second equity capital flow variable is obtained by the following adapted version of Dichev‟s (2007) measure of net equity capital distributions:
DISTRit,,it-x = Ln[MVit-x * (1+ rit-x,it)] – Ln(MVit) (3)
Where DISTRit,,it-x measures total equity capital distributions for firm i between month t-x and
month t, MVit is the market value of firm i at time t, and rit,it-x is the simple stock return for
firm i between time t-x and time t. The period buy-and-hold return (BHR), or simple stock return, for a firm is computed as:
𝑟𝑖𝑡 , 𝑖𝑡−𝑥 =𝐼𝑖𝑡−𝐼𝑖𝑡 −𝑥
𝐼𝑖𝑡 −𝑥 (4)
where 𝐼𝑖𝑡 and 𝐼𝑖𝑡 −𝑥 are closing total return indices of firm i at at time t and t-x and rit, it-x is the
BHR of firm i over the period from time t-x through time t. For monthly returns x = 1 and for yearly returns x = 12.
16
A.1.2. Independent Variables Past and future abnormal returns
I calculate abnormal returns for each firm for the one- and three-year period prior to (AR-12,0 and AR-36,0), as well as for the one- and three-year period after (AR12,24 and AR12,48) the year of equity distribution measurement. Firm simple returns are calculated by applying equation (4). Abnormal returns are generated by deducting from each firm‟s simple return the contemporaneous return on a UK market index, which is taken to be the MSCI U.K. index.
Book-to-market equity
I calculate book-to-market ratios for each firm as the book value of equity (WC03501) at the end of a month, divided by the contemporaneous market value of equity (MV). As some book-to-market values appear to be erroneous at first sight – some observations being extremely high, others being very close to zero – I proceed by winsorizing monthly book-to-market values at the 1st and 99th percentile for every firm in the sample. A variable BM, which is constructed as the natural logarithm of the book-to-market value for the end of December, the previous year, is used to forecast aggregate equity distributions starting January the next year until January the year thereafter. I also create a dummy variable, BMdum, in order to preserve degrees of freedom in equation estimation. In cases where the book value of equity is unavailable or negative, I set both the book-to-market variable and the book value dummy variable equal to zero. The dummy variable takes a value of one when the book-to-market variable is available and positive.
Market value
The monthly market value of equity (MV), calculated as share price multiplied by number of shares outstanding, is used as a control variable. I refer to the natural logarithm of the market value at the end of June as MV, and use it in regressions to predict equity distributions from the beginning of the current year‟s July through the end of June the next year.
B. Data Summary and Model Estimation B.1. Data Summary
17
from regression estimation turn out non-normally distributed (Brooks (2008) pp. 164). The capital flow measures are of particular interest here: despite the fact that they both measure log changes in capital, the measure of yearly share issuance (ISSUE0,12) and that of yearly equity distribution (DISTR0,12) are strongly skewed,exhibiting skewness measures of 5.07 and
Table I
Descriptive Statistics: Equity Capital Flows and Proxies for Market Timing Opportunities Reported variables are measure for future year total equity issuance: a, ISSUE0,12 = [Ln(adjusted shares, t + 12) – Ln(adjusted shares, t)]; and another variable measuring total equity distributions, which is adapted from Dichev (2007): DISTR0,12 = Ln[MV0* (1+ r0,12)] – Ln(MV12); the natural logarithm of the ratio of the book value of equity to the market value of equity, measured at the end of year t-1, BM; the natural logarithm of firm market value at the end of year t-1, MV; and one- and three-year future as well as past abnormal returns. Abnormal returns are created by deducting contemporaneous returns on the MSCI U.K. index from firm returns. The variables are from monthly measures, for 250 large U.K. firms that have been active throughout the sample period from January 1985 through the end of 2009.
A) Summary statistics ISSUE0,12 DISTR0,12 BM MV AR-12,0 AR-36,0 AR12,24 AR12,48 Mean 0.05 -0.01 1.85 4.94 0.05 0.22 0.05 0.13 Median 0.00 0.03 1.85 4.79 0.00 0.01 0.00 -0.04 St. dev. 0.22 0.23 0.14 2.30 0.45 1.52 0.45 1.05 Skewness 5.07 -4.42 -1.85 0.25 3.42 18.03 3.30 4.23 Kurtosis 82.77 49.96 18.10 2.61 38.62 715.94 36.51 56.68 Obs. 72,250 75,000 65,976 75,000 74,912 74,260 75,000 66,000 B) Correlation matrix
Variable ISSUE0,12 DISTR0,12 BM MV AR-12,0 AR-36,0 AR12,24
DISTR0,12 -0.90 BM 0.01 -0.01 MV -0.12 0.14 -0.35 AR-12,0 0.03 -0.04 0.07 0.01 AR-36,0 0.02 -0.04 -0.10 0.05 0.41 AR12,24 -0.03 0.04 0.09 -0.05 -0.03 -0.08 AR12,48 -0.05 0.06 0.09 -0.07 -0.07 -0.11 0.54
18
procedure may be the cause of „survivorship biased‟ results, something which must be taken into account when research inferences are drawn.
Panel B presents the contemporaneous correlations between the variables. The measure of firm equity distributions DISTR0,12 is highly negatively correlated with aggregate equity issuance ISSUE0,12 (ρ = -0.90). This is as expected since the variables measure almost the same capital flows, but in opposite directions. None of the independent variables has a stronger correlation than 0.35 (between MV and BM) with any of the other variables. I therefore do not expect any serious multicollinearity problems in regression estimation.
B.2. Fama-MacBeth Cross-Sectional Regressions
Following the methodology Pontiff and Woodgate (2008) use in their research of the relationship between equity distributions and future returns, I estimate separate regressions for each month of data in the vein of Fama and MacBeth (1973) and then calculate the average of the regression coefficients for all months in the sample. Panel data can be subject to two general forms of dependence that are important to consider when performing regressions. Firstly, the residuals of a given period may be correlated across different firms (cross-sectional dependence). And second, the residuals of a given firm can be correlated across periods (time series dependence). Fama-MacBeth standard errors, by themselves, are robust to cross-sectional dependence, but not to time series dependence. (Petersen (2009)). To take account of any heteroscedasticity as well as the possibility of serial correlation, I use the variance-covariance estimator developed by Newey and West (Brooks (2008) pp. 152). Again following Pontiff and Woodgate (2008), I estimate each regression with 11 lags in the Newey-West procedure3, 11 being equal to the number of months over which equity distributions are calculated minus 1. The frequency of the data is thus used to decide the truncation lag length.
C. Results
For six different model specifications, average slope coefficients and intercepts as well as average R2s are reported in table II. This table shows the expected coefficient sign reversals when going from panel A – where equity issuance is the dependent variable – to panel B – where equity distribution is the dependent variable – and the estimated coefficients in the two panels do not differ much in significance. Looking at the differences in average R2-values between the two panels, the independent variables appear to be slightly better at explaining
3
I estimate these regressions in Stata 11, using an ado-file for Fama-MacBeth estimation with Newey-West corrected errors, obtained from Judson Caskey‟s website:
19
firm equity distribution (DISTR0,12) than equity issuance (ISSUE0,12). For the sake of brevity, I only discuss the results for the regressions with ISSUE0,12 as the dependent variable in detail here. Interpretations for the regressions with DISTR0,12 (panel B in table II) can be derived from this discussion in an obvious manner.
The first four rows of panel A in table II represent a „horse race‟ to determine the extent to which, respectively, book-to-market (BM), firm market value (MV) and past- and future abnormal returns can explain yearly aggregate equity issuance. Only the signs for the one- (AR12,24) and three-year (AR12,48) future abnormal stock return turn out as expected. The coefficient sign for past abnormal return (AR-12,0 and AR-36,0) is actually opposite from what was expected, and book-to-market (BM) on its own appears to have no significant effect on equity issuance. The relation found between market value (MV) and equity issuance is significantly negative with a convincingly large t-statistic (t = -7.45), a finding consistent with the explanation that smaller firms, which presumably have more future growth opportunities, issue more equity capital.
When all three mispricing proxies and the control variable (MV) are used simultaneously in one regression estimation, as in the last two rows of panel A, past abnormal returns no longer appear to have a significant effect on equity issuance, with t-statistics between -0.3 for the three-year time horizon (R-36,0) and -1.4 for the one year horizon (R-12,0). On the other hand, the negative relationship between issuance and future abnormal returns remains statistically significant (t < -3 for both the three-year and the one-year time horizon). The coefficient on BM has the expected sign and is marginally significant in these specifications ( -2.33 < t < -1.87). The likely cause of this last observation is the inclusion of MV and BM in the same specification, since these two variables are significantly negatively related with a correlation coefficient of -0.35 (see table I).
20
Table II
Fama-MacBeth Regressions: Predicting Equity Capital Flows
Fama-MacBeth cross-sectional regressions with Newey-West corrected standard errors are performed with monthly data. Two different measures are used as dependent variables: ISSUE0,12 = [Ln(adjusted shares, t +12) – Ln(adjusted shares, t)]; and DISTR0,12 = Ln[MV0* (1+ r0,12)] – Ln(MV12). The measures for equity capital flows are regressed on the following variables: the natural logarithm of the ratio of the book value of equity to the market value of equity, measured at the end of year t-1, BM; a dummy variable, taking the value 0 when the book-to-market ratio is not available and 1 when it is available, BMdum; the natural logarithm of firm size at the end of year t-1, MV; and one- and three-year future as well as past abnormal returns. Abnormal returns are created by deducting contemporaneous returns on the MSCI U.K. index from firm returns. The results presented are the average regression coefficients of 300 Fama-MacBeth regressions with Newey-West corrected t-statistics in parentheses. The Newey-West procedure uses 11 lags. The regressions are for 250 large U.K. firms that have been active throughout the sample period of 25 years from 1985 through 2009, with a total of 75,000 firm-month observations. All coefficients are multiplied by 100. Coefficients that are significant at the 5% level are bolded.
A) Dependent variable is 1-year future aggregate equity issuance (ISSUE0,12)
Intercept BM BMdum MV AR-12,0 AR-36,0 AR12,24 AR12,48 Avg. R2
15.36 (7.88) 1.36 (0.67) -14.35 (-3.61) 3.10 10.85 (9.93) -1.32 (-7.45) 2.04 4.06 (9.37) -2.00 (-1.45) -3.61 (-3.94) 2.10 4.44 (12.43) -1.32 (-2.03) -2.39 (-3.90) 2.03 19.64 (8.65) -4.64 (-1.87) 1.36 (-0.30) -1.24 (-6.58) -0.58 (-0.56) -3.40 (-4.18) 6.38 21.03 (9.14) -6.31 (-2.14) 1.56 (0.30) -1.41 (-8.19) -0.72 (-1.40) -2.05 (-4.26) 6.84
B) Dependent variable is 1-year future aggregate equity distribution (DISTR0,12)
Intercept BM BMdum MV AR-12,0 AR-36,0 AR12,24 AR12,48 Avg. R2
21
size (as proxied by market value), which is used as a control variable – without any hypothesized sign – in the model specifications, appears to have a stronger ability to predict aggregate equity issuance than any of the proxies for market timing opportunities. Lastly, there appears to be only a marginally significant negative relationship between book-to-market and aggregate equity issuance, after correcting for firm size.
D. Conclusion
Let us return to the question this section aims to answer: “Can we observe U.K. firms attempting to time the equity market?”. Considering the results presented above, the answer is not entirely straightforward. Not all three of the mispricing-proxies appear to be significantly related to aggregate equity flows. One- or three-year past stock returns do not help explain equity issuance decisions, and the book-to-market ratio has a minor impact, if any. The only mispricing proxy that is consistently significantly related to cross-sectional equity issuance is the future abnormal stock return, whether measured over one or three years. But if the correlation between issuance and future returns alone constitutes evidence that managers are exploiting market timing opportunities, this would imply that managers are able to correctly predict an underreaction of the stock market to their issuance (distribution) decision. In order to do so, managers would need to have private information about the true value of the firm, i.e. they must perceive the firm to be misvalued by the market (Myers and Majluf (1984)). But if this is true, the under- or overvaluation they perceive is not revealed by past stock returns and/or the book-to-market ratio in the manner the over-extrapolation theory of DeBondt and Thaler (1987) and Lakonishok et al. (1994) predicts – in which case there would be a clear „window of opportunity‟ for equity market timing. I therefore consider a reverse causation more likely, namely, that the equity issuance decisions are not taken on the basis of correct predictions of future returns, but that the abnormal performance somehow results from the equity distribution decisions. All in all, the findings in this part of the study provide little support for the hypothesis that U.K. managers have been purposefully attempting to time the equity market.
IV. Managerial Market Timing Performance
22
there are cross-sectional determinants of market timing success. The results indicate that, whereas there is a large variability of market timing performance across the sample, the average firm in the sample has not been successfully timing the equity market. Market timing ability appears to be positively related to firm size, but I find no relations between market timing success and firm liquidity, book-to-market ratio or the industry a firm operates in.
A. Firm Performance Gaps
To be able to answer this paper‟s second sub question, I need a measure of market timing that is comparable across firms. The methodology used by Dichev (2007) – introduced in the literature section – seems ideally suited, since it gives us a point estimate of managerial timing ability: the „performance gap‟. First, Dichev (2007) shows that net capital distributions between firms and investors can be estimated as:
𝑁𝐷𝑖𝑡 = 𝑀𝑉𝑖𝑡 −1∗ 1 + 𝑟𝑖𝑡 − 𝑀𝑉𝑖𝑡 (5)
where 𝑁𝐷𝑖𝑡 measures net distributions for firm i in period t in currency units, 𝑀𝑉𝑖𝑡 is market capitalization, and 𝑟𝑖𝑡 is the return for to stock i in period t. Positive distributions are equity capital flows from the firm to investors and negative distributions are flows from investors to the firm. Measured in this way, the flows include dividends, share repurchases and stock issues but also less straightforward items such as the exercise of stock options, issuing stock in mergers and acquisitions and contributions and distributions of noncash assets.
The measure which according to Dichev (2007) gives the actual return earned by investors, the dollar-weighted period return (DWR), is calculated as the internal rate of return (IRR) where the market capitalization of a firm at the beginning of the period enters with a negative sign, total monthly distributions for the firm enter with their sign, and total ending market capitalization of the firm enters with a positive sign. This procedure comes down to solving the following formula for 𝑟𝑖𝑑𝑤:
𝑟𝑖𝑑𝑤: 𝑀𝑉 𝑖0 1 + 𝑟𝑖𝑑𝑤 𝑇 − 𝑁𝐷𝑖𝑡 1 + 𝑟𝑖𝑑𝑤 𝑇−𝑡 = 𝑀𝑉 𝑖𝑇 𝑇 𝑡=1 (6)
23
From the firm‟s point of view, the difference between the buy-and-hold return and contemporaneous dollar-weighted return (i.e. BHR – DWR), the performance gap, can be interpreted as a point estimate of the reduction in the cost of equity due to successful market timing. If the firms in the sample have issued and repurchased equity capital randomly across time, their dollar-weighted returns will not differ much from buy-and-hold returns. But if issues, repurchases and mergers have purposefully been timed by managers, we expect to find significant positive gaps on average. With Dichev‟s (2007) method I calculate monthly equity distributions (equation (5)), buy-and-hold returns (4), dollar-weighted returns (6) and performance gaps for each firm in the sample. I do this for the entire sample period and separately for two sub-periods, 1985-1997 and 1998-2009. I overcome the critique Keswani and Stolin (2008) raise with respect to Dichev‟s (2007) international tests (i.e. those for markets outside the U.S.), since I look at individual firm returns and market capitalization values, not at Datastream index figures which are sensitive to changes in the database‟s coverage. Table III reports a summary of my findings. The table shows that the average performance gap for the sample firms, as calculated over the entire sample period, is actually
Table III
Buy-and-Hold Returns and Dollar-Weighted Returns for the Sample Firms
This table gives summary statistics for the 250 firms in the sample. Buy-and-hold returns, dollar-weighted returns, and gaps are calculated for each firm over the whole sample period and over two sub-periods. Buy-and-hold return is the annualized geometric average of monthly simple returns. Dollar weighted return is the annualized IRR from a calculation in which initial market value enters with a negative sign, monthly distributions enter with their sign, and ending market value enters with a positive sign. The performance gap is equal to firm buy-and-hold return minus dollar-weighted return.
Statistic Average Stdev Median Min Max
24
a negative 1.4%. The median performance gap is higher, a negative 0.5%. We can conclude that the 250 prominent U.K. firms in this sample have over the past 25 years, on average, not been successful at timing the equity market. That said, the performance gaps do show a large range, a left skew and a relatively high standard deviation of 6.2% across the sample, indicating that some of the firms in the sample were successful nonetheless. The two sub-periods similarly show negative average performance gaps, albeit very small ones (-0.3% for the first and -0.2% for the second sub-period). Gaps for the sub-periods also display high variability across the sample, with a standard deviation of 7.1% for the years 1985-1997 and 6.7% between 1998 and 2009. The finding of dissimilar average gap sizes when gaps are measured over different time periods is in line with Keswani and Stolin (2008), who use examples to show that because internal rates of return are nonlinear in nature, dollar-weighted return measures are very sensitive to aggregation across time periods and markets. In short, the performance gap-measure of market timing performance only gives a point estimate over the time horizon measured, and it can therefore only be interpreted as the average market timing performance of a firm over the time period chosen.
B. Market Timing Performance Across the Sample
In the previous subsection I calculated performance gaps for every individual firm in the sample. But this paper aims to determine the role of market timing in the U.K. stock market as a whole, in which large firms have more impact than small ones. In order to determine market timing success across the sample, I follow Dichev‟s (2007) research more closely by constructing a value-weighted index of the sample firms. Table IV shows yearly summary statistics for this index. The monthly aggregate distribution of equity capital to investors in the indexed firms is calculated as aggregate market value of the ordinary shares at the end of the previous month, cumulated at the value-weighted return for that month, less the aggregate market value at the end of the month. Yearly equity distributions are equal to the sum of monthly aggregate distributions. The buy-and-hold return for the index of firms is equal to the annualized geometric average value-weighted return, given by:
𝐵𝐻𝑅 ≡ 𝑤𝑖 𝑇𝑖 1 + 𝑟𝑖𝑡 − 1
𝑡=1 𝑁
𝑖=1 (7)
Where 𝑟𝑖𝑡 is the return for to stock i in period t and 𝑤𝑖 is stock i‟s weight in forming the average period holding return, 𝑤𝑖 = 𝑀𝑉𝑖 𝑀𝑉, where 𝑀𝑉𝑖 is the common stock market value
25
Table IV
Descriptive Statistics for a Value-Weighted Index of the Sample Firms
Panel A presents summary statistics for an index consisting of the 250 firms in the sample. Market cap is the aggregate year-end market value (share price*number of shares outstanding). Distributions is the sum of total yearly net pound distributions of all firms. Monthly distributions are computed as NDit = MVit-1 *(1 + rit ) – MVit
where NDit measures net distributions for security i in period t, MVit is market capitalization, and rit is the total
return (including dividends) to stock i in period t. Annual return is the compounded measure of monthly value-weighted return for the year; this return measures the experience of investors who, at the beginning of each month readjust the weights of the equities in their portfolios according to the market capitalization of the equities. Market cap is aggregate year-end market capitalization. Relative distribution is total net distributions for the year divided by the average of beginning and ending market capitalization. Panel B reports correlations between yearly relative equity distributions and past- and future yearly value weighted return, with t-statistics in parentheses.
A) Summary statistics
Year Market Cap
(£ million)
Return Total Distributions (£ million) Relative Distrs. 1985 97,723 0.256 1,047 0.012 1986 118,534 0.269 -434 -0.004 1987 124,081 0.167 10,218 0.079 1988 143,948 0.093 2,054 0.015 1989 185,043 0.405 6,505 0.038 1990 179,792 -0.049 8,689 0.046 1991 212,866 0.238 4,343 0.022 1992 246,169 0.207 7,171 0.031 1993 304,691 0.259 -1,841 -0.006 1994 294,476 -0.047 7,571 0.025 1995 359,069 0.243 5,610 0.017 1996 415,857 0.175 3,728 0.009 1997 500,098 0.269 1,162 0.002 1998 553,495 0.119 -23,263 -0.040 1999 710,488 0.209 -2,018 -0.003 2000 757,550 -0.033 -97,403 -0.124 2001 710,255 -0.075 39,273 0.051 2002 581,950 -0.186 32,344 0.050 2003 613,220 0.172 25,520 0.043 2004 670,034 0.130 19,043 0.029 2005 790,097 0.222 30,458 0.041 2006 838,971 0.130 42,106 0.050 2007 857,527 0.033 35,314 0.041 2008 542,820 -0.260 48,035 0.066 2009 753,885 0.245 -25,586 -0.038 Mean 0.127 7,186 0.018 Stdev 0.159 28,621 0.041
*Performance gap measured over the index: -0.004
B) Correlations
Relative Distr. Future return
26
An examination of table IV reveals that the aggregate market capitalization of the sampled firms has generally increased throughout the sample period, from about £100 billion in 1985 to £750 billion at the end of 2009. This increase is mainly caused by firm stock returns, since equity capital distributions have been positive on average. That is, cash flows from firms to equity investors have outweighed flows in the other direction by – on average – a little more than £7 billion per year. Value-weighted returns averaged 12.7% over the sampled years, with a standard deviation of 16%. These figures are in line with existing evidence on historical stock returns. Equity distributions, when scaled by aggregate market capitalization, show high variability over the years, with a standard deviation of 3.1% at an average of 1.8% per year. Furthermore, relative equity distributions display similar swing-patterns to those Dichev (2007) encounters for NASDAQ-listed firms, showing mainly positive distributions but some periods of more frequent negative distributions, especially during 1998-2000. What specifically stands out is the large negative distribution of close to £100 billion in the year 2000. Dichev (pp. 393) finds a similarly large negative distribution in the U.S. in the same year. In the U.S., this negative distribution is followed by negative stock returns in the magnitude of more than 30%, suggesting that investors poured in huge amounts of money at a very wrong time. But for the sample of U.K. firms, the return subsequent to this large negative distribution is a mere negative 3%. In light of this, is not very surprising that I find the correlations between relative distributions and past- and future yearly index returns, as reported in panel B of table IV, to be insignificantly different from zero. Of course, these correlation coefficients are calculated over a small sample with only 25 data points, but their insignificance provides no preliminary support for the hypothesis that the aggregate dollar-weighted return differs from the buy-and-hold return, thus that successful market timing has taken place.
27
C. Firm Characteristics and Market Timing Performance
The findings reported in table III of subsection A above indicate the presence of a large variability in firm market timing performance, as measured by the performance gap. Even after finding that the average firm in the sample actually lost money – by distributing equity capital at the wrong time – I still wonder what sets apart those firms with higher gaps. Can we distinguish firms with differing degrees of market timing ability? Indeed, it seems plausible that of those firms that attempt to time the equity market, some profit more than others. In this subsection I explore whether there are factors that can distinguish those firms with higher performance gaps. To this end, firm gaps, as measured over the whole sample period, are regressed on the following firm-characteristic variables: market value, book-to-market ratio, relative trading volume and dummies for firm industry.
C.1. Independent Variables
28
the industry a firm operates in, I expect that financial firms are better at timing the equity market for the simple reason that they are specialized in financial services. It is my presumption that finance experts perform better with financial transactions than the average corporate manager.
I have only one performance gap available per firm with which to perform tests, and I must therefore rely on a central tendency measure – in this case the median observation over the sample period – of each of the variables to serve as regressors.
Table V
Market Timing Performance in the Cross-Section of Firms
Panel A gives summary statistics for the firms in the sample. Data were collected monthly for each firm, from January 1985 to December 2009, so the statistics are calculated over 75,000 firm-months. Reported variables are: the performance gap, measured as BHR-DWR; the logarithm of the median market value of the firm over the sample period ln(med.MV); the logarithm of median relative monthly trading volume, measured as trading volume divided by contemporaneous market value, ln(med.Rel.Vol), and the logarithm of the median book-to-market ratio ln(med.BM). Panel B shows the industries in which the firms in the sample operate according to their ICB industry codification. Panel C shows the correlations between the variables from panel A. Panel D gives the coefficients of cross-sectional OLS regressions with HAC-consistent standard errors. Dependent variable is firm performance gap and independent variables are the variables from panel A and four industry dummies. The regressions are performed on 250 cross-sectional observations. All coefficients are multiplied by 100; t-statistics are reported in parentheses. Coefficients that are significant at the 5% level are bolded.
A) Summary statistics
Variable Gap Ln(med.MV) Ln(med.Rel.Vol) Ln(med.BM)
Mean -0.01 5.01 -4.14 -0.50 Median -0.00 4.78 -4.11 -0.55 St. dev. 0.06 2.18 0.27 0.66 Skewness -2.02 0.34 0.84 0.17 Kurtosis 11.19 2.43 5.74 3.54 Obs. 250 250 250 247 B) Benchmark industries
Industry N.o. firms
29
C) Correlation matrix
Variable Gap Ln(med.MV) Ln(med.Rel.Vol)
Ln(med.MV) 0.24
Ln(med.Rel.Vol) 0.14 0.53
Ln(med.BM) -0.10 -0.42 -0.22
D) OLS cross-sectional regressions with HAC-consistent standard errors: dependent variable is firm performance gap
Intercept Ln(med.MV) Ln(med.Rel.Vol) Ln(med.BM) Adj. R2 -5.00 (-4.83) 0.72 (3.98) 6.07 9.70 (0.95) 2.86 (1.10) 1.02 -1.79 (-3.58) -0.94 (-1.72) 0.64 -2.51 (-0.25) 0.63 (2.41) 0.49 (0.22) -0.01 (-0.02) 4.61
Intercept Industrials Cons. Goods Cons. Services Financials Adj. R2 -1.11
(-1.26) (-0.27) -0.28 (-0.60) -0.75 (0.07) 0.09 (-0.46) -0.68 -0.01
Panel A in table V below shows summary statistics for the variables. The dependent variable – performance gap – has a leptokurtic distribution with a strong left skew (excess kurtosis = 8.19; skewness = -2.02). Unfortunately, I see no easy way of transforming this variable so that its distribution becomes more symmetric. The logs of median firm market value and book-to-market ratio both show a close to normal distribution, with a minor right skew of 0.34 and 0.17 and an excess kurtosis of -0.57 and 0.54 respectively. Median relative trading volume follows a slight leptokurtic distribution with a minor skew to the right. Panel B lists the industries in which the sample firms operate. The majority of firms are from the industrial, consumer services, financial or consumer goods sector (96, 45, 39 and 34 firms, respectively)4. Panel C presents correlation coefficients between the variables. Market value and relative trading volume appear to be strongly positively correlated, with a coefficient of 0.53. This observation indicates that even when trading volume is corrected for differences in firm market value, larger firms still tend to have more liquid stock. There is a negative correlation between book-to-market and both market value and relative trading volume, with
4
30
respective correlation coefficients of -0.42 and -0.1. The two correlations with an absolute magnitude higher than 0.4 may cause multicollinearity problems in OLS regression estimation. I take this into account by estimating separate regressions for each of the independent variables in the next subsection.
C.2. Regression Estimation and Results
The results of five separate cross-sectional OLS regressions, with heteroscedasticity and autocorrelation consistent (HAC) standard errors, are reported in panel D of table V. The signs of the coefficients on firm market value and relative trading volume are positive – as expected – but only market value appears to significantly affect the performance gap. When the gap is regressed on median market value alone, the slope is 0.72% with a t-statistic of 3.98 and a reasonable level of explained variation (measured by adjusted R2)of 6.07%. The coefficient sign when regressing on book-to-market ratio alone is negative but insignificant (t = -1.72). When firm gap is regressed on market value, book-to-market and relative volume jointly, the signs of the coefficients do not change, but their significance declines. The regression of firm performance gap on 4 different industry dummies is what we may call a „junk regression‟: all dummy-coefficients are far from significant and the adjusted R2
takes on a negative value. We may therefore conclude that the differences in performance gap cannot be explained by the industry in which firms operate.
I thus find indications that market timing success may be related to firm size, larger firms being better at timing the equity market than smaller ones. Furthermore, market timing performance appears unrelated to trading volume, book-to-market and operating industry. That said, I want to be careful in drawing inferences here. The tests in this subsection are performed on only 250 observations, which is probably not enough to appeal to the central limit theorem and assume that the test results will asymptotically follow appropriate distributions (Brooks (2008), pp. 164)5. Furthermore, the tests rely on central-tendency measures, and even though the dependent variable – firm performance gap as defined by Dichev (2007) – is a theoretically and logically appealing measure of market timing success, it does not (as pointed out earlier in this section) seem to be a very stable measure.
D. Conclusion
This section aims to answer a question that consists of two parts: a) To what extent are U.K. firms successfully timing the equity market? And b) What distinguishes those firms that
31
are (more) successful at timing the equity market? Regarding the first part, I find that the average firm in my sample of U.K. firms was – during the period from January 1985 to December 2009 – not successful at timing the equity market. When the sample period is subdivided into a twelve-year and a thirteen-year period, the average firm appears not successful during either of these periods. Similarly, a value-weighted index constructed of the sample firms shows no evidence of market timing across the sample. On the other hand, the measure for market timing success, the performance gap that can be calculated for each firm in the sample, displays a high variability across the sample. This implies that some firms in the sample may have indeed, whether purposely or not, been successfully timing the equity market. Regressions of firm performance gaps on different cross-sectional characteristics indicate that firm performance gap is positively related to firm size, whilst it is not related to operating industry, trading volume or book-to-market. Thus, regarding the second part of the question, larger firms seem more successful, or in this case, „less unsuccessful‟, at market timing.
V. Equity Capital Flows and Future Stock Returns
This section researches whether equity capital flows provide extra information to predict future stock returns. The equity market timing hypothesis predicts that positive equity distributions are followed by negative long-run abnormal returns, and vice versa. There exists a large body of literature on long-run returns following equity capital flows (i.a. Ritter (1991), Loughran and Ritter (1995), Ikenberry, Lakonishok and Vermaelen (1995), Pontiff and Woodgate (2008)). The main result in these papers – and what I find here – is that on average, equity issues are followed by low returns and repurchases are followed by high returns.
A. Regression Variables
32
As in the section on equity distributions and market timing opportunities, I conduct tests both with Pontiff and Woodgate‟s (2008) composite issuance measure and the adapted version of Dichev‟s (2007) total equity distribution measure (defined by equation (3)). I use both annual equity capital flow measures, following Pontiff and Woodgate (2008), as well as three-year measures, since most studies into long-run returns after equity flows find a significant relationship between issues or repurchases and three-year future returns (i.a. Ritter (1991), Loughran and Vijh (1997, Chan, Ikenberry and Lee (2007), Ang, Gu and Hochberg (2007), Ishikawa and Takahashi (2011)). I disregard the use of the five-year variable exemplified by Daniel and Titman (2005), since my sample period is relatively short in comparison to theirs (they research more than 35 years of data).
In addition to the composite share issuance variable, „ISSUE‟, and the adapted measure of Dichev‟s (2007) total distribution variable, „DISTR‟, I use firm market-to-book equity, firm size and a measure for momentum as independent variables in the regressions. These variables are chosen because they have been found to be associated with stock returns in past research. Specifically, firms with lower market-to-book ratios and smaller size have been shown to have higher stock returns on average (see Fama and French (1992); Lakonishok, Schleifer and Vishny (1994); and Hyde and Sherif (2010) for specific U.K. evidence). Past short-term stock returns have been shown to be positively related to future returns (Carhart (1997)); this pattern is captured by the momentum measure. For the reader‟s convenience, the variables that are used in this section are not only described below, but also listed in appendix B.I., along with their definitions.
A.1. Variable Measurements A.1.1. Dependent Variable Stock returns
I perform various regressions with stock returns over different holding periods as the dependent variable. The stock returns are all calculated according to equation (4): 𝑟𝑖𝑡, 𝑖𝑡−𝑥 =(𝐼𝑖𝑡 − 𝐼𝑖𝑡 −𝑥) 𝐼𝑖𝑡 −𝑥; where 𝐼𝑖𝑡 and 𝐼𝑖𝑡 −𝑥 are closing total return indices of firm i at at time t and t-x, and x is measured in months. To avoid extreme observations having an undue weight in the regression estimations, I winsorize the monthly return variables for each firm, setting the lowest 1% of the observations equal to the 1st percentile and the highest 1% of the observations to the 99th percentile.
33
again, in light of the existing evidence of long-run (that is, three-to-five year) returns following equity distributions.
A.1.2. Independent Variables Equity issuance/distribution
The measure of aggregate equity issuance as defined by Pontiff and Woodgate (2008) and described by equation (1) is measured over a one-year as well as a three-year horizon (so x=12 in the first and x=36 in the second case when applying equation (1)). I name these variables ISSUE1 and ISSUE3 respectively. Both variables are expected to be negatively related to future stock returns, if equity market timing is at stake. In cases where the three-year issuance variable is not available, I make use of a dummy variable, called ISSUE3dum, in order to preserve degrees of freedom in equation estimation. When ISSUE3 is not available, I set both ISSUE3 and ISSUE3dum equal to zero. Otherwise, ISSUE3dum takes on a value of one.
As in section IV, I also – separately – test with „aggregate equity distribution‟, the measure defined by equation (3), that also captures dividends, on top of stock issues, repurchases and stock mergers and acquisitions. Like in the case of aggregate issuance, this measure is calculated over the one- as well as the three-year time horizon. I name these variables DISTR1 and DISTR3 respectively, and seeing that they measure capital flows in the opposite direction of the issuance measure, I expect them to be positively related to future returns. As for the issuance measure, I also create a dummy variable DISTR3dum, which is set to zero when DISTR3 is unavailable.
Book-to-market equity
