1.7.1 Event Studies from Zero

Basic Prerequisites: Econometrics & Time Series (return aggregation) and Probability & Measure Theory (expectations).

---

1. Intuition & Practical Objective

This page builds the why of event studies with no prior event-study knowledge needed. The objective is one idea: to measure how a specific corporate event moved a stock price, you cannot just look at the price move - you must subtract what the stock would have done anyway, on that day, for reasons unrelated to the event. That subtraction is the entire method, and everything else is refinement.

Start with the dumbest question: why can’t we just read the +2% jump on the announcement date? Because on any given day a stock also moves for the market (the whole index rose or fell), for its sector, for general news, and for plain noise. A stock can gain 2% on the announcement day for no event reason at all - because the market rallied 1.5% that day. So the event’s true contribution is the residual: observed return minus the return the stock “should” have earned given everything other than the event.

Three steps:

1. Expected return = the “no-event” counterfactual. Define a normal (expected) return - what the stock would earn if the event had not happened. The simplest version (Brown & Warner’s mean-adjusted model) uses the stock’s own average daily return in a quiet estimation window before the event. Abnormal return = observed − that average.

2. Abnormal return is a wealth effect. The abnormal return at the event is a direct measure of the unanticipated change in securityholder wealth caused by the event (Kothari–Warner 2007). A +3% abnormal return says “this event transferred ~3% of firm value, over and above market/risk effects.” Because rational investors price news instantly, most of the effect shows up on the announcement day itself - which is why the method works even with a window of one to three days.

3. Aggregate across firms to cancel noise. One firm’s abnormal return is dominated by idiosyncratic noise (a single +3% event on a 2%-volatility stock is swamped). Average the abnormal returns over many firms that shared the same type of event; the noise averages toward zero and the common signal survives. This cross-sectional aggregation is the statistical heart of the method.

---

2. Mathematical Ground Truth & Derivations

The return decomposition (Kothari–Warner 2007, eq. 1–2). Write the observed return as a normal component plus an abnormal component:

$$R_{it}=K_{it}+e_{it},$$

where $K_{it}$ is the expected (normal) return under a given model and $e_{it}$ is the abnormal return. Equivalently,

$$e_{it}=R_{it}-K_{it},$$

the difference between the return conditional on the event and the expected return unconditional on the event.

Mean-adjusted model (Brown–Warner 1980/85, eq. 1–2). The normal return is the stock’s own estimation-window average:

$$K_{it}={\bar{R}}_i=\frac{1}{L}\sum _k{R}_{ik},A{R}_{it}=R_{it}-{\bar{R}}_i,$$

over an estimation window (BW use days $-244$ to $-6$, ~239 observations) chosen before the event so it cannot be contaminated by the event itself.

Why “abnormal return” is a test of efficiency. If markets are semi-strong efficient, the abnormal return should be nonzero only in the narrow window around the announcement and zero afterward. Systematically nonzero abnormal returns after the event are inconsistent with efficiency - they would support a profitable (pre-cost) trading rule (Kothari–Warner 2007, §3.2.2). This is exactly the logic that makes event studies the “cleanest evidence we have on efficiency” (Fama 1991).

---

3. Computational Implementation - the simplest abnormal return

This is the entire method in four lines - the mean-adjusted model from Brown & Warner. Stdlib only.

The firm moved +2.5%, but once we subtract its usual −0.22% baseline, the abnormal return is +2.72% - slightly larger, because the firm happened to trend down when it “should” have been flat. The event’s true, unanticipated wealth effect is the residual, not the raw move.

---

4. Failure Modes & First-Principles Breakdowns

1. The “raw-return” trap. Looking at the raw price move instead of the abnormal return mistakes market drift and firm beta for event impact. Brown–Warner explicitly flag a “Raw Returns” variant (mean-adjusted with ${\bar{R}}_i$ forced to zero) that finds spurious abnormal performance when the average return is positive - because it ignores the baseline the stock would have earned anyway.
2. The estimation window must be clean. If the estimation window (before the event) itself contains value-relevant events, the baseline ${\bar{R}}_i$ is biased and every abnormal return is off by that bias.
3. One firm tells you nothing. A single +3% abnormal return on a 2%-volatility stock is statistically indistinguishable from noise. The method only works when abnormal returns are averaged over a cross-section of firms sharing the event type - see 02 · Methodology.
4. Expected-return model is a choice. “Abnormal” is only defined relative to a model of normal returns. A different benchmark (mean-adjusted vs market model vs Fama–French) gives a different abnormal return - event studies are joint tests of the event and the benchmark (see 04 · Statistical Testing).

---

5. References

• Brown & Warner (1980), Measuring Security Price Performance, JFE 8(3)
• Brown & Warner (1985), Using Daily Stock Returns: The Case of Event Studies, JFE 14(1)
• Kothari & Warner (2007), Econometrics of Event Studies, Handbook of Corporate Finance
• Fama (1991), Efficient Capital Markets: II, J. Finance 46(5)

---

6. Connected Graph Bridges

• Base: Econometrics & Time Series · Probability & Measure Theory
• Continue: 02 · Event-Study Methodology · Index Hub
• Sibling: Cross-Sectional Momentum (averaging a signal across many assets - the same aggregation logic)