Changes

Method of payment measures (cash vs. stock) are most popular, and present in half of all papers covered. Other transaction characteristics are rarely used. Acccounting-based measures, particularly Tobin's Q, and Analyst Forecast measures, particularly the Std. Dev. of Forecasts, are the next most popular and occur in approximately 1/3rd of all papers covered. Price/Volumet Volume measures, particularly the idiosyncratic volatility, are the next most popular, occuring in approximately 1/5th of all papers covered. ==Rejecting Variables== The following variables are explicitly or implicitly related to the CAR in an acquisition and so unsuitable:*Pre-CAR*Abnormal/Unexpected Turnover*Momentum - It is unclear from a quick read of Tetlock (the only paper that uses this measure) whether momentum is an artifact of information asymmetry or a response to it's mitigation. *Stock Illiquidity - see Momentum above. The following are simply too hard to get on a reasonable timescale:*News Announcements*Ratings Changes*Target Age (not in SDC, so we would have to get another source...)*R-Squared from Earnings, Book Value - this was used in a single paper and isn't worth it (it requires joining CRSP to COMPUSTAT before running the regressions) ==Build Notes== ===Idiosyncratic Variability=== We want <math>\sigma_{\epsilon}^2</math> from <math>R_it = \alpha + \beta_i R_mt + \epsilon</math> run annually for each publicly-traded firm in the NYSE/Nasdaq/Amex universe. This is equivalent to the RMSE as: <math>\mathbb{E}(\epilson) = 0 \quad \mathbb{V}(\epilson)= \mathbb{E} \left( (R-\hat{R} - (\alpha-\hat{\alpha}) - (\beta - \hat{\beta})R_m \right)^2 = RMSE^2</math> Data:*Annual data from CRSP*Draw entire universe (>2Gb?)*Rely on date stamps*Use CRSP Permo (or Cusip?) - Don't need to draw NAICS if we are going to join back...*Holding Period Return*Value-Weighted Return inc. distributions Run the regressions on raw data (i.e., don,t join to COMPUSTAT first). ===Ratio of Shares Traded=== Defined as: "the ratio of number of shares traded during the last year ending before the equity issue announcement, divided by the number of shares outstanding at the end of the fiscal year before the ... announcement." We can compute it on an average over an annual basis using CRSP quarterly ("Shares Traded" isn't in COMPUSTAT), or using CRSP daily (same data as above), either way we want: Data:*Share Volume (VOL)*No. of Shares Outstanding (SHROUT)*And to take an average over the year for each firm. ===Analyst Forecasts=== At least one paper reported problems with the data before 1991 (see the lit review). *Forecast Error (Analysts over and under react): <math>ForecastError=\frac{|ACT_t-EST_t|}{|Act_t|}</math>*Std. Deviation of forecasts (Correlated with riskiness): <math> ForecastSD=\frac{SD_t}{|Act_t|}</math>*Range of Forcasts*No. Estimates*No. Analysts*Normalized forecast error: <math>NormForecastError=\frac{ForecastError}{\sigma_{ACT_t - ACT_0}</math> I.e., Over some time period calculate the detrended variation in Earnings. This probably isn't worth it.From KS99: "the normalized forecast error, which is defined as the ratio of the forecast error in earnings to the earnings volatility of the firm. Earnings volatility is the standard deviation of the firm's detrended quarterly earnings in the five-year period before the announcement of the spin-off." Data:*From I/B/E/S Detail file pull:*CUSIP (8Dg)*EPS*Fiscal Yr1*Analyst Code*Estimate Value*Actual Value Match back to COMPUSTAT to get NAICS.