Empirical Methods in Corporate Finance

Size: px

Start display at page:

Download "Empirical Methods in Corporate Finance"

Allan Ward
5 years ago
Views:

Close ties between accounting research and corporate finance.

1 Uses of Accounting Data Josh Lerner Empirical Methods in Corporate Finance Accounting-based Research Why examine? Close ties between accounting research and corporate finance. Numbers important to both. Use many of the same data sets. Focus on many of the same issues. Asymmetric information. Agency costs. Monitoring. 1

2 Accounting-based Research Information is central to accounting research. Asymmetric. Bias. Accuracy. Choice. Accounting as a representation of economic value. Accounting-based Research Little communication between the disciplines. Few cross-over publications. Little cross-referencing to similar topics. Today meant to be introduction. Lots of scope for diving deeper. 2

3 Types of Accounting-based Research Today. Earnings management. Accounting policy choices. Analyst forecasts. Others. Accounting policies. Earnings-based methods of valuation and returns. This session Detecting manipulation. Measurement error: Industry categorization. Implications for stock prices. 3

4 Detecting Earnings Management Dechow, Sloan, and Sweeney AR, 1995 Deliberate misstatements Firms will misrepresent their performance: Firms will misrepresent their performance: Revenue manipulation (e.g., channel stuffing) Expense manipulation (e.g., capitalizing operating expenses; warranty reserve; postemployment benefit assumptions; fictitious transactions with related parties) 4

5 But challenging to detect Earnings management cannot be detected with certainty. Tests that identify such behavior may also identify certain firms with shared characteristics. If related to accounting performance, inferential challenges. Particularly problematic in studies of consequences of earnings management: What if tests of earnings management identify firms undergoing multi-year changes? What if tests of earnings management identify attributes that are priced in market? Focuses on identifying earning management To what extent do existing models succeed in capturing earnings management, avoiding false identifications? Modeled after Brown and Warner [1980, 1985]: Simulations to assess probability of Type I, II errors. Also using cases where enforcement actions brought. 5

6 Definitions Accrual: Provisions for future positive or negative future cash flows when cash has not yet been transferred : Often scaled by assets, e.g.: TA t = (( Δ CA t Δ Cash t ) ( Δ CL t Δ STD t ) DEPR t ) / A t 1 Discretionary accrual: Similar to unexpected component of stock return. Typical earnings management study Actual estimated relationship: Regress discretionary accruals on dummy denoting events suspected of having earnings management: D.A. Proxy Would like to estimate: = a + b PARTITION t + e t D. A. = α + βpartition t + γ X + ε K k = 1 k kt t 6

7 Two essential problems Cannot observe discretionary accruals. Cannot explain all factors that lead to discretionary accruals. Consequences: Incorrect attributions of earnings management (Type I error). Failure to identify cases of real earnings management, due to biased coefficients or inflated standard errors (Type II error). Different models for identifying discretionary accruals Healy [1985] model: Average of total accruals/lagged assets over an estimation period. Unbiased if accruals move randomly around constant mean. DeAngelo [1986] model: Total accruals/lagged assets in previous year. Unbiased if accruals follow random walk. 7

8 Different models (2) Jones [1991] model: Adds control for economic changes. Estimates coefficients from regression in estimation period: TA t / At = a1(1/ At 1) + a2( ΔREVt / At 1) + a3( PPEt / At 1) + υ Generates predicted ditdchange in earnings. Modified version uses change in revenues net change in receivables: Assumption is that credit sales may be managed. t Different models (3) Industry (Dechow-Sloan [1991]) model: Regress in estimation period: TA t / At = γ 1 + γ 2MedianIndustry ( TAt / Ay Coefficients, contemporaneous industry values estimate predicted non-discretionary accruals. ) 8

9 Experiment Examine: 1000 random firm-years firm-years where in top or bottom deciles in earnings or cash flow random firm-years with extra earnings : Delayed expense recognition. Accelerated revenue recognition. Margin manipulation (costs and revenues increase). 56 firm-years where SEC actions. Results All do well explaining normal case. Jones model does better in years with extreme earnings: All have problems. When induce earnings management, Jones and Industry models do best. In actual cases, similar results. 9

10 Take-awaysaways Identifying earnings management is non- trivial. More complex models that incorporate more information do best. Must be exceedingly careful, particularly in settings with outlying earnings. Diversification Discount or Premium? Villalonga JF

11 Motivation Work on diversification discount showing that valuation of diversified firms is lower Increasing number of papers questioning the result. Bias in who diversifies. Campa and Kedia (2002). Motivation (2) Segment data is very poor. Historically, have used COMPUSTAT selfreported segment data. Required to report segments that are 10% or more of the company. The most number of segments is 10. These segments can be aggregated and actually cover many different types of activities. 11

12 Motivation (3) Sometimes firms change segment reporting with no real change in the business: IBM to typewriters and other. Data Business Information Tracking Services. g Census data at the establishment level. Has information on revenues, assets, employees, SIC code, payroll, location, etc. at the establishment level. Sample. All those on both COMPUSTAT and BITS No survivorship bias. 12

13 Advantages of BITS Disaggregated. Establishments are linked longitudinally. Covers all sectors of economy. Census Data in General Census data is great for looking at public and private firms. Lots of interesting questions to ask. Have to use census data on their machines, i.e., can t download data to your computer, registration process, etc. But much easier with NBER data center. 13

14 Summary Stats Table I A lot more business units than segments. Business unit is aggregated establishments at the four digit SIC code level within a given firm. Table II sample. A lot more firms with a lot more business units than segments reported on COMPUSTAT. Table IV. Matched q of industries for business units is higher. Approach Follow the Lang and Stulz methodology. Look at industry matched q taken from single segment firms and average by revenue, employees, etc. based on the segment size. Look at true single segment firms. Far more single segment firms if you define it using COMPUSTAT. (~60% are single segment) Get the q for the actual firms and then get the sales weighted average q for the individual segments. Table VI look at excess value for BITS and Compustat companies. 14

15 Robustness Checks Look at different weights. Look at various sub samples. Restricting to those that match at the 2 digit SIC or finer. Table VII. Table VIII look at different measures of diversification and excess value. Consistent result With COMPUSTAT segment data, get a discount. With BITs data, get a premium. What Explains the Result? Related explanation Related diversification may be good. Can only look at BITS because relatedness can be observed. Strategic Accounting Firms want to hide how well they are doing. Hide good performing groups. Manipulate data to show poor performance. 15

16 Conclusion Better data makes a difference. Looking at establishment level data is better proxy of diversification. Diversified firms seem to trade at a premium, not a discount. Not certain of cause of premium. The relationship between corporate financing activities, analysts forecasts, and stock returns Bradshaw, Richardson, and Sloan JAE

17 Motivation Ritter (1991) and Loughran and Ritter (1995) show LR underperformance after stock issuance Spiess and Affleck-Gras show underperformance after bond issuances Can authors use analyst forecast behavior to test theories Possible Explanations Risk Misvaluation Wealth Transfer 17

18 Measures of External Finance Utilize Compustat annual files ΔXFIN=ΔEQUITY+ΔDEBT Equity is net cash from sale of stock less cash dividends Acquisitions? Debt includes converts, sub debt, notes, debentures, and capitalized leases scale everything by average assets. Dealing with Extreme Outliers Need robust regressions Winsorizing Censoring Robust regressions Here winsorize if absolute value of change greater than 1 18

19 Measuring Performance Use only size-adjusted returns Why no Book-to- Market? Calculate 12 month returns Analyst Forecast Errors Take Actual earnings minus forecast, Negative numbers imply optimistic forecast Do for 1 year, 2 Year, long term growth J target price, and investment recommendation 19

20 Tables 1 and 2 Sample More 1 and 2 year forecasts Summary stats Returns test Rank firms on three external finance measures and put into deciles Do BHAR using size matched portfolios No BTM or skewness adjustments Hedge is the return of long low external finance short high external finance Table 3 and Figure 1 20

21 Calendar time results Table 4 Do annual CAPM and FF regressions Use the annual BHR in regressions the EW Questions why annual? why only EW? why no momentum factor? Is this just the (1,1) again? Fama-McBeth Regressions Run annual cross sections Estimate average coefficient and t-stat from the time series of coefficients Still only site adjusted returns Table 5 21

22 Look at Relation to Analyst Forecast Errors Table 7 and Figure 2 Tests Difference in means and regressions Tables 8 and 9 Figure 3 Look at affiliated vs. unaffiliated analysts No diff Conclusions External finance predicts returns Controls? Not state of the art External finance related to forecast errors No relation to affiliated analysts Conclude that it is misvaluation 22

23 Overall Assessment Lots of interesting data in accounting. Similar problems. Room for substantial collaboration. Research opportunities Other biases are out there! Industry segmentation is doubtless just one of many. Opportunity to look carefully at methodologies that take for granted and see flaws. 23

24 Research opportunities (2) Methodological re earnings management: Can press farther into sorting between unexpected accruals and managed ones? Certain subsets of balance sheet items? More analyses in the spirit of Dechow, et al. s [1995] final analysis? Will these give different answers? Research opportunities (3) Exploration of real effects of earnings management farther: What are reputational consequences? E.g., reactions to securities issues? To what extent do these activities affect: Market share? Innovation? etc.? 24

Analysis on accrual-based models in detecting earnings management

Lingnan Journal of Banking, Finance and Economics Volume 2 2010/2011 Academic Year Issue Article 5 January 2010 Analysis on accrual-based models in detecting earnings management Tianran CHEN tianranchen@ln.edu.hk