earnings announcements and price behavior sam lim
TRANSCRIPT
Earnings Announcements and Price Behavior
Sam Lim
A Little Background Information Content of Earnings
Announcements Beaver 1968
Landsman and L. Maydew 2002 Abnormal volatility Volatility increases around quarterly earnings
announcements
Kinney et al 2002 “Surprise” materiality in returns
Most surprises and returns are of same sign, but 43 to 45% of firms’ surprises associated with returns of opposite sign
S-shaped surprise return relation
Use HAR-RV as suggested
Summary of last time HAR-RV
Earnings surprise factor (percentage)
( EPSactual - EPSestimate ) / EPSactual * 100
Split-sign regression Statistically significant positive findings
Surprise correlated with increase in volatility Not too surprising.
(continued) Standardize returns, as suggested by Dr.
Tauchen Return divided by square root of RV
Alison Keane finds weekly RV works relatively better than daily or monthly RV, so I follow suit
Mostly same result–surprise often correlated with overnight returns, but not intraday returns. Previously, had a problem—surprise sometimes correlated with intraday returns. Turns out F-statistic is very low in those cases, so cannot reject null hypothesis that surprise does not determine direction of intraday returns. Price corrections happen fairly quickly, before market
open
Not all firms of same interest E.g. Goldman Sachs GS
32 positive surprises, 3 negative surprises, 1 hits estimate (no surprise). Not very interesting.
Positive surprises positively correlated with volatility at .1% level, positively correlated with overnight returns at 1% level, no correlation with intraday returns.
McDonald’s 14 positive surprises, 10 negative surprises, 19
hits estimate
McDonald’s 4/9/1997 to 1/7/2009 14 positive, 10 negative, 19 no surprise Did not account for quarters when firm just
hits estimate (SURPRISE=0), so generate dummy variable for earnings release with no surprise.
Generate dummy variables for positive and negative days as well, for comparison purposes. May be better to do anyway?
MCD HAR-RV regression, omit RV1, RV5, RV22 for
simplicity.
All significant at .1% level
Nice results? Fits with theory that negative news has more impact on the market than positive news.
No surprise days also increase volatility! Why?
Analyst estimates discounted? Dispersion of estimates? Hopefully not, but could also be release of other news on same
day.
RVt Positive Negative No surprise
Coefficient 3.13 4.23 2.12
Another look – Merck, UPS, Pepsi MRK - 17 positive surprises, 7 negative, 19 no surprises
Positive significant at 5% level, negative and no surprise significant at 1% level
UPS - 19 positive, 5 negative, 9 no surprises
All significant at .1% level
PEP – 24 positive, 8 negative, 11 no surprises
Positive and no surprise significant at .1%, negative at 1%
Though positive coefficient is larger…
RVt Positive Negative No surprise
Coefficient 1.06 2.58 1.24
RVt Positive Negative No surprise
Coefficient 2.04 3.54 3.09
RVt Positive Negative No surprise
Coefficient 2.03 1.83 3.45
Accounting for dispersion? Account for dispersion in analyst estimates, as
suggested by Dr. Bollerslev The less consensus among analysts, the less
information the market has (mean estimate has less informative value)
Interaction term created with standard deviation of analyst estimates
No surprise days (using McDonald’s)
No surprise significant at 5%, dispersion at .1%, interaction at 10%
RVt No surprise
Dispersion
Interaction
Coefficient 1.97 201.5 -183.8
Dispersion, continued Makes more sense using absolute value of
surprise, but this begs the question of whether I should use the surprise value, or the dummy values.
All statistically significant at .1% level.
RVt |SURPRISE|
Dispersion
Interaction
Coefficient 1.03 284.35 -103.82
Another issue: Sub-sampling
Walmart: Importance of sampling rate 28 positive, 6 negative, 10 no surprise
Sampled at 15 minutes
Positive significant at 5%, no surprise significant at 10%
Sampled at 10 minutes
Negative significant at 10%
Problematic?
RVt Positive Negative No surprise
Coefficient 1.03 Not significant
1.24
RVt Positive Negative No surprise
Coefficient Not significant
1.66 Not significant
Further work Have a list of different S&P 100 firms, is there
some systematic pattern to results? Industry? When looking at returns, picture further
complicated. Do announcements of one firm affect another
firm’s stock behavior? Jumps?
Last time, concluded jumps do not occur in higher frequencies on earnings release dates. Perhaps this is not the case? IBM – 27.9% jumps (BNS test at 5% level) on earnings
release dates, 16.3% other days.. Similar numbers for Intel.