Working Paper: CEPR ID: DP18524
Authors: Ian Martin; Ran Shi
Abstract: We introduce a framework that uses option prices to deliver upper and lower bounds on the probability of crash in an individual stock, and argue based on a priori considerations that the lower bound should be close to the true crash probability. Empirical tests support this prediction in and out of sample. We horse-race the lower bound against a range of characteristics proposed by the prior literature. The lower bound is highly statistically significant, with a t-statistic above five, and is an order of magnitude more economically significant than any of the characteristics, in the sense that a one standard deviation increase in the lower bound raises the predicted probability of a crash by 3 percentage points, whereas a one standard deviation change in the next most important predictor (a measure of short interest) moves the predicted probability of a crash by only 0.3 percentage points.
Keywords: forecasts
JEL Codes: G12; G17
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| risk-neutral probability of a crash (D81) | upper and lower bounds (C62) |
| lower bound outperforms lasso-based competitor (C51) | predicting crash probabilities (G17) |
| lower bound on crash probability (C62) | actual crash occurrences (G01) |
| lower bound (D20) | realized crash indicators (G01) |
| lower bound (D20) | predict crash probabilities (C53) |