Earnings Surprises and Signal Analysis matt mc Connell

Earnings Surprises and Signal Analysis matt mc. Connell David Nabwangu Eskil Sylwan Johnson Yeh

Agenda Background n Hypothesis n Methodology n Data Fitting n Explanatory Variables n Regression Results n Conclusion n

Market Reaction to News We expect stock price to react to news like this: an Reality Ideal World In a. In More Realistic World Delay Overshoot Settling Time Announcement Date

Step Response of a Second. Order System +- Parameter Effect Initial Level before reaction begins Offset Beginning of reaction Magnitude Size of reaction ωn ωd ζ Shape of curve

Hypothesis 1. Abnormal returns after earnings surprises follow a curved pattern which can be modeled using the step response of a second-order system 2. 6 Curve parameters are predictable using information about the company * Results could be applicable to any news item – earnings are measurable

Methodology 1. Identify earnings surprises (Factset) • 2. 3. 4. Retrieve price and other data series (Datastream) Calculate abnormal returns in ± 30 day window Fit a curve to each event • • 5. 600 events, 100 companies Least squares method with solver 6 parameters for each event Regress 6 parameters on several explanatory variables

Fitting the Data In some instances the data fit very well Correlation = 91. 6% Average Correlation = 80% In some instances fit not good Correlation = 70%

Explanatory Variables 6 explanatory variables for curve parameters 1. Quarterly Earnings Surprise % 1. Positive influence on magnitude, Zeta 2. 1 -Year Price Growth 1. Negative Impact on Offset, Positive Impact on Magnitude, and Zeta 3. Quarterly Earnings Surprise $ 1. Positive Impact on Magnitude, Zeta 4. Price to Earnings Ratio 1. Positive Impact on Offset, Negative Impact on Zeta 5. Beta 1. Positive Impact on wm, wd, & Magnitude 6. 10 -Day Abnormal Return 1. Negative Impact on Magnitude

Regression Results

Conclusion n Holds promise: Some predictive power n Paths forward: – Better fitting method n Least squares method more applicable to linear – Improve predictive regressions n n More predictor variables Non-linear predictor variables – Test predictability over time – Larger data set – Create and test trading strategies