Working Paper: NBER ID: w26765
Authors: Tobias Salz; Emanuel Vespa
Abstract: We evaluate dynamic oligopoly estimators with laboratory data. Using a stylized en-try/exit game, we estimate structural parameters under the assumption that the data are generated by a Markov-perfect equilibrium (MPE) and use the estimates to predict counterfactual behavior. The concern is that if the Markov assumption was violated one would mispredict counterfactual outcomes. The experimental method allows us to compare predicted behavior for counterfactuals to true counterfactuals implemented as treatments. Our main finding is that counterfactual prediction errors due to collusion are in most cases only modest in size.
Keywords: dynamic oligopoly; collusion; Markov-perfect equilibrium; experimental economics
JEL Codes: L10; L13
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| Violation of the Markov assumption (C69) | Biased parameter estimates (C51) |
| Violation of the Markov assumption (C69) | Prediction errors in counterfactual scenarios (C53) |
| Collusion (D74) | Downward biased estimates of model parameters (C51) |
| Collusion (D74) | Inaccurate predictions (C52) |
| Collusion affects individual choices (D70) | Overall dynamics of the market (E32) |
| Breakdown of collusion (D74) | Reversion to MPE patterns (C59) |
| Frequency of successful collusion attempts (D74) | Reasonable approximation of the Markov assumption (C60) |