Measuring Price Impact with Investors’ Forward-Looking Information

The recent Supreme Court decision in Halliburton brought renewed interest to price impact and event studies. Aside from identification and analysis of the news itself, the event study has three basis steps: (i) Estimate a statistical model (or “market model”) of how the stock price would be expected to change in absence of such news (“predicted price changes”), (ii) Calculate stock price changes in excess of the predicted price changes (“excess price change”), and (iii) Evaluate the statistical significance of the excess price change to distinguish material news from noise, or normal variations in stock prices.

A market model describes how a stock price behaves under normal circumstances—its typical levels of variability and how much it fluctuates with market or industry benchmarks. Market models are intended to be forward-looking, putting into statistical terms how market investors would expect a stock to behave on average. Market models are typically based on a statistical sample of market data, either from a time period preceding the news events to be tested or a time period surrounding the events (with statistical controls to prevent the model from being influenced by the events themselves).

However, market models based on samples of historical data make a critical assumption that the market conditions and investor expectations from the sample period apply equally to events being tested. This is not necessarily the case, particularly if the sample data covers a time period of changing market conditions in the stock being examined or the overall market. For example, a market model estimated prior to the financial crisis might not be appropriate to testing events during the crisis itself. As another example, a market model estimated prior to a large merger might be mis-specified as compared to one that would reflect the impact of the merger on the company’s stock price.

We introduce a market model that infers investor expectations from market prices as of the event itself rather than from a sample of data spanning some other time period. This way our model reflects investors’ forward-looking expectations as of the time of the events being tested and avoids the tradeoffs involved in choosing among different sample periods. This approach can be preferable to the traditional approach because it substitutes objective market-based criteria for subjective expert judgment, at the same time more closely tying the event study method to market information.

More specifically, our method uses information embedded in stock option prices to estimate market models, implement event study statistical analyses, and derive conclusions on price impact. Such a forward-looking methodology can produce results that diverge from those of traditional approaches. Both the magnitude of excess returns and their statistical significance can differ from the results obtained through traditional approaches. For example:

• We tested an alleged disclosure event in a securities suit using both approaches. Traditional approaches found the disclosure event to be statistically significant. However, those methods ignored the increase in uncertainty reflected in investor expectations of stock volatility. By using a forward-looking market model we found that the disclosure was not significant at the 95% confidence level.
• We evaluated the price impact of a major investment bank announcing an equity sale at the height of the financial crisis. Forward-looking market models showed a substantially more positive price impact than using traditional methods.

By incorporating contemporaneous information, the forward-looking market model can improve the relevance and reliability of event studies. We view this method as another tool in the expert’s toolbox: it does not necessarily replace the traditional event study methods, but instead provides additional analytical possibilities in evaluating competing expert opinions, providing corroborative evidence to traditional methods, or implementing event studies in situations where they are limited by the quantity or relevance of sample data.

Additional information is available here. Detailed technical information is available in our research paper available here.