ANALYSIS TESTS 4.1 INTRODUCTION
4.3 OHLSON (1995) VALUE RELEVANCE MODEL REFORMULATION
4.3.4 Regression Model Estimation
Time series estimation of regression models (15) to (23) is conducted using Ordinary Least Squares (OLS) estimation. The time series standard error estimates are based on Newey-West heteroskedasticity and autocorrelation consistent standard errors to overcome the problem of non-constant variance and autocorrelation of error terms, a problem that is especially important in regression equations (15) to (20) where the share price dependent variable (Pt+1) will be highly persistent. We also obtain coefficient estimates using fixed time effects, fixed firm effects, and pooled estimation.12 These latter (non-time series) coefficient standard error estimates are based on White’s heteroskedasticity-consistent standard errors to overcome the problem of non-constant variance of the cross-sectional error terms. For comparison purposes with Jennings, LeClere, and Thompson (2001), each estimated regression equation is assessed using
11 We also check the sensitivity of the results to the use of a 12-month (instead of a three month) price
change dependent variable in regression equations (21) to (23), and the results remain unchanged.
12 The pooled samples do not contain the same number of observations each year because of missing
observations in some of the firm time series. Details for the fixed effect coefficient estimation are provided in the results tables.
adjusted R2, in addition to assessing the statistical significance of the coefficient estimates. [Note, however, that Brown, Lo, and Lys (1999) indicate that adjusted R2 is not an appropriate measure for assessing the explanatory power of value relevance regression models, due to scale effects whereby the scale (or size) of dependent and independent variables in value relevance studies affects the apparent explanatory power of the models.]
4.4 DATA
The data set is obtained from the United States COMPUSTAT database. The data set consists of quarterly equity price data (DATA14) and annual earnings-based data. The annual variables are earnings per share before extraordinary items (DATA58), intangible assets (DATA33), amortisation of intangibles (DATA65), goodwill (DATA204), amortisation of goodwill (DATA394), and number of common shares outstanding (DATA25).
The earnings per share data have been manipulated to satisfy the data requirements for our study, as in Jennings, LeClere, and Thompson (2001). Firstly, goodwill amortisation is estimated when it is not directly reported.13 Goodwill amortisation per share (GAPS) is determined as goodwill amortisation (DATA394) divided by shares outstanding (DATA25).14 Earnings per share are then adjusted to
13 The Financial accounting Standard Board has implemented two new accounting standards for goodwill
accounting (SFAS 141: Business Combination, and SFAS 142: Goodwill and Other Intangible Assets) effective from financial year 2002. Under the new standards, firms no longer account for goodwill amortisation in their financial statements. Firms are allowed, however, to provide goodwill amortisation information separately with other financial information.
14 Goodwill amortisation is estimated in accordance with the method devised by Jennings, LeClere, and
Thompson (2001): (1) directly reported amortisation of goodwill (GWA) is directly used. Otherwise, (2) if current year goodwill (GW) equals current year intangible assets (IA) then the amortisation of goodwill (GWA) equals amortisation of intangibles (IAA), i.e., if GW=IA then GWA = IAA; (3) if GW≥0, IAA≥0, and IA=0 or missing (“ ”), then GWA = IAA; (4) if GW>0.9*IA (i.e., >90% of GW), then GWA =
obtain earnings per share before goodwill amortisation (EBG) and earnings per share after goodwill amortisation (EAG).15 The quarterly and annual datasets are merged based on classifications common to both datasets.
Our study examines a 16 year period, 1988 to 2003, when goodwill amortization was potentially reported.16 To conduct the time series analysis, sample firms must report earnings for a minimum of 12 years as well as have positive estimated goodwill amortisation for a minimum of nine years, thus avoiding domination by zero goodwill amortisation values. In the sample period 374 firms report earnings for at least 12 (75%) of the 16 years, and 58 of these 374 firms have positive estimated goodwill amortisation for at least nine years. A choice is made to analyse a randomly selected sample rather than the entire 58 firms. We therefore randomly select 20 firms (out of the 58 firms with sufficient data) to illustrate the benefit of including the most recent prior period’s equity price in the time series analysis.17 The full names of the firms are provided in Panel A of Table 4.1, along with the symbol used to designate the firms in the results tables. Summary statistics for the data set as well as a correlation table for the data set variables are provided in Table 4.1. The pooled descriptive measures and percentile measures for market equity value (MEV) are also reported to indicate that the company time series sample represents random sampling of both small and large firms (see Panels B and C (IAA*GW)/IA; and (5) if GW<0.9*IA and 0.9*GWL<GW<GWL, then GWA = GWL-GW, where GWL = last (previous) year goodwill.
15 Because of new accounting rules (SFAS 141: Business Combination, and SFAS 142: Goodwill and
Other Intangible Assets) introduced by the Financial Accounting Standards Board (FASB), DATA58 (EPS – earnings per share) is reported in COMPUSTAT in two ways: before 2002 as after goodwill amortisation, and from 2002 as before goodwill amortisation. For the years 2002 onwards, we then adjust DATA58 (earnings per share) to include goodwill amortisation in order to obtain earnings after goodwill amortisation (EAG). For the years before 2002, DATA58 is adjusted to exclude goodwill amortisation in order to determine earnings before goodwill amortisation (EBG).
16 Goodwill amortization is no longer reported in COMPUSTAT from 2004 onwards, so 2003 is the
endpoint of the data sample.
17 We do not restrict the analysis to certain industries since, unlike in other corporate finance studies,
there is no a priori reason why the relationship between share price (or price change), earnings, and goodwill amortisation should differ between industries, and even if it did then the fixed firm effect estimation would account for this.
of Table 4.1). Panel D of Table 4.1 reveals that the most recent prior period’s equity price (Pt) is highly correlated with current trailing earnings per share (EBG or EAG), thus revealing that current trailing earnings could act as a proxy for the most recent prior period’s price if the prior period’s price is not included in value relevance regression analysis.18
[Please insert Table 4.1 about here.]