Quarterly Earnings Releases, Expectations, and Price Behavior Sam Lim
Set-up Purpose: to explore the relationship between analyst expectations, quarterly earnings releases, and stock price behavior. Analyst earnings estimates and actual earnings obtained from Wharton’s WRDS, from the I/B/E/S database. Release time of quarterly earnings announcement (BMO or AMC) obtained from Earnings. com
Set-up (continued) HAR-RV Model RV is annualized Earnings surprise factor (percentage) ( EPSactual - EPSestimate ) / EPSactual * 100 Run HAR-RV adding the surprise factor as a regressor. On days of quarterly earnings announcements (the day after, if announcements are made AMC), SURPRISE = surprise factor. Otherwise, SURPRISE = 0.
Chevron (CVX) Prices sampled every 10 minutes Data from 10/10/2001 to 01/07/2009 (1804 days), 26 quarterly earnings releases (BMO) RVt+1 RVt-5, t RVt-22, t SURPRISE constant Coeff. 278. 584. 073 -. 044. 349 Std. Error. 029. 044. 034. 023. 094 P-value 0. 000 0. 033 0. 062 0. 000
Chevron (CVX) Split-sign regression – Are the effects of negative surprises different from positive surprises? CVX – 12 positive surprises, 13 negative surprises RVt+1 RVt-5, t RVt-22, t SURPRISE(+) SURPRISE(-) constant Coeff. 277. 583. 075. 010 -. 077. 338 Std. Error. 029. 044. 038. 030. 094 P-value 0. 000 0. 029 0. 787 0. 010 0. 000
Amazon (AMZN) Prices sampled every 5 minutes Data from 08/01/1997 to 01/07/2009 (2846 days), 42 earnings releases (AMC), 14 positive surprises, 23 negative surprises Surprise days not properly lagged RVt+1 RVt-5, t RVt-22, t SURPRISE(+) SURPRISE(-) constant Coeff. 301. 302. 337. 024 -. 010. 749 Std. Error. 022. 038. 035. 022. 018. 218 P-value 0. 000 0. 294 0. 601 0. 000
Amazon (AMZN) Surprise days lagged one day to account for AMC announcements RVt+1 RVt-5, t RVt-22, t SURPRISE(+) SURPRISE(-) constant Coeff. 300. 302. 337. 027 -. 046. 749 Std. Error. 022. 038. 034. 022. 018. 218 P-value 0. 000 0. 238 0. 012 0. 001
Pepsi (PEP) Prices sampled every 5 minutes Data from 04/09/1997 to 01/07/2009 (2925 days), 66 earnings releases (BMO), 24 positive surprises, 8 negative surprises RVt+1 RVt-5, t RVt-22, t SURPRISE(+) SURPRISE(-) constant Coeff. 313. 287. 315. 120 -. 091. 491 Std. Error. 022. 038. 035. 041. 106. 112 P-value 0. 000 0. 003 0. 390 0. 000
Chevron (CVX) Overnight returns ln(price at market open) – ln(price at market close from previous day) ONReturn SURPRISE(+) SURPRISE(-) constant Coeff. 00019. 00050. 00031 Std. Error. 00016. 00015. 00021 P-value 0. 247 0. 001 0. 140
Chevron (CVX) Intraday Returns Sum of returns within the day Return SURPRISE(+) SURPRISE(-) constant Coeff. 00030. 00017. 00031 Std. Error. 00027. 00024. 00021 P-value 0. 261 0. 473 0. 783 Increase in volatility not from everyone selling after negative surprises…
Chevron (CVX) BNS Jump Test (Quad Power, Ratio-max adjusted) Percentage of Jump days No big difference – lagging has no real results either volatility increase not from jumps Though there could be intraday jumps… BNS Z-Score All days Earnings Rel. days Not ER days . 1% 1. 94 0 1. 91 1% 5. 88 3. 85 5. 74 5% 14. 36 15. 38 14. 23
Amazon (AMZN) Overnight Returns ONReturn SURPRISE(+) SURPRISE(-) constant Std. Error. 00027. 00024. 00052 P-value 0. 000 0. 007 0. 994 Intraday Returns Return SURPRISE(+) SURPRISE(-) constant Coeff. 00062. 00026. 00000 Coeff. 00027 -. 00004. 00071 Std. Error. 00016. 00013. 00072 Sign-split regression oddities? P-value 0. 096 0. 762 0. 319
Pepsi (PEP) Similar results as Amazon regressions. Regressing overnight returns with surprise – statistically significant, positive relationship (p-value is nearly 0) Regressing intraday returns with surprise – statistically insignificant, slightly negative relationship(p-value. 15) However, split-sign regression yields positive relationship significant at 10% level, but only for positive surprises (not stat. sig. for negative surprises)
Pfizer (PFE) Prices sampled every 5 minutes Data from 04/09/1997 to 01/07/2009 (2923 days), 43 earnings releases (BMO), 31 positive surprises, 6 negative surprises RVt+1 RVt-5, t RVt-22, t SURPRISE(+) SURPRISE(-) constant Coeff. 313. 287. 315. 120 -. 091. 491 Std. Error. 022. 038. 035. 041. 106. 112 P-value 0. 000 0. 003 0. 390 0. 000
Bank of America (BAC) Prices sampled every 15 minutes Data from 04/09/1997 to 01/07/2009 (2923 days), 42 earnings releases (mostly BMO), 31 positive surprises, 7 negative surprises RVt+1 RVt-5, t RVt-22, t SURPRISE(+) SURPRISE(-) constant Coeff. 339. 434. 174. 070 -. 430. 320 Std. Error. 022. 034. 027. 063. 034. 089 P-value 0. 000 0. 264 0. 000
Further analysis Try Lee-Mykland test for jumps, to see if there are intraday jumps occurring. Account for dispersion in analyst expectations. Try to find what is the norm/exception (Chevron has the nicest results, is the norm or exception? ) If Chevron’s results are the norm, how long does this uncertainty after a earnings surprise last? Incorporate other stock-specific news announcements to see effect on stock price behavior (similar to Alison Keane’s research on macroeconomic news announcements) Continuing with the effects of analysts theme, perhaps look at analyst recommendations (buy/hold) and stock behavior.
Скачать презентацию Quarterly Earnings Releases Expectations and Price Behavior Sam
fd77aabd3839d4e643ea1e36016237b1.ppt