An Easier Way to Calibrate

@article{Fudenberg1999AnEW,
  title={An Easier Way to Calibrate},
  author={Drew Fudenberg and David K. Levine},
  journal={Games and Economic Behavior},
  year={1999},
  volume={29},
  pages={131-137}
}
Forecasts are said to be calibrated if the frequency predictions are approximately correct. This is a refinement of an idea first introduced by David Blackwell in 1955. We show that “ K -initialized myopic strategies†are approximately calibrated when K is large. These strategies first “initialize†by making each forecast exactly K times, and thereafter play, in each period t , the minmax strategy in a static game. 
Forecast Hedging and Calibration
TLDR
The concept of forecast hedging is developed, which consists of choosing the forecasts so as to guarantee that the expected track record can only improve, and which yields all the calibration results by the same simple basic argument.
Mostly calibrated
TLDR
A prequential test based on calibration that also fails the informed forecaster on at most a category I set of true distributions and is non-manipulable is presented, which sheds light on the relationship between likelihood and calibration with respect to the distributions they reject.
How to Buy Advice
A decision maker, whose payo is inuenced by an unknown stochastic process, seeks the advice of an advisor, who may be informed about the process. We identify a sucient condition on the correlation
Good sequential probability forecasting is always possible
Building on the game-theoretic framework for probability, we show that it is possible, using randomization, to make sequential probability forecasts that will pass any given battery of statistical
Comparative Testing of Experts
We show that a simple "reputation-style" test can always identify which of two experts is informed about the true distribution. The test presumes no prior knowledge of the true distribution, achieves
The Principal-Agent Approach to Testing Experts
TLDR
It is shown that there often exist contracts that allow a decision maker to attain the first-best payoff without learning the expert's type, and the problem of testing probabilistic models into one of taking advantage of information that could be contained in experts' advice.
Good randomized sequential probability forecasting is always possible
Summary.  Building on the game theoretic framework for probability, we show that it is possible, using randomization, to make sequential probability forecasts that will pass any given battery of
A Geometric Proof of Calibration
TLDR
This proof follows from a direct application of Blackwell's approachability theorem to a carefully chosen vector-valued payoff function and convex target set and captures the essence of existing proofs based on approachability.
Calibrated Forecasting and Merging
Consider a general finite-state stochastic process governed by an unknown objective probability distribution. A forecaster, observing the system, assigns subjective probabilities to future states.
Any Inspection is Manipulable
CONSIDER A SEQUENTIAL FORECAST that may vary with time and history. A calibration test (see Dawid (1982), Foster and Vohra (1997), and Kalai, Lehrer, and Smorodinsky (1999)) compares the observed
...
...

References

SHOWING 1-10 OF 18 REFERENCES
The Well-Calibrated Bayesian
Abstract Suppose that a forecaster sequentially assigns probabilities to events. He is well calibrated if, for example, of those events to which he assigns a probability 30 percent, the long-run
Calibrated Forecasting and Merging
Consider a general finite-state stochastic process governed by an unknown objective probability distribution. A forecaster, observing the system, assigns subjective probabilities to future states.
Conditional Universal Consistency
Players choose an action before learning an outcome chosen according to an unknown and history-dependent stochastic rule. Procedures that categorize outcomes, and use a randomized variation on
A simple adaptive procedure leading to correlated equilibrium
We propose a new and simple adaptive procedure for playing a game: ‘‘regret-matching.’’ In this procedure, players may depart from their current play with probabilities that are proportional to
A Proof of Calibration Via Blackwell's Approachability Theorem
Over the past few years many proofs of calibration have been presented (Foster and Vohra (1991, 1997), Hart (1995), Fudenberg and Levine (1995), Hart and Mas-Colell (1996)). Does the literature
Self-Calibrating Priors Do Not Exist
TLDR
It is shown that self-calibrating priors in the sense of Dawid (1982a) do not exist.
An analog of the minimax theorem for vector payoffs.
for all i, j . Thus in the (two-person, zero-sum) game with matrix Λf, player I has a strategy insuring an expected gain of at least v, and player II has a strategy insuring an expected loss of at
Comment: The Impossibility of Inductive Inference
Oakes's argument shows that no statistical analysis, however complex, of sequential data can be guaranteed to provide asymptotically valid forecasts for every possible set of outcomes.
Asymptotically Subminimax Solutions of Compound Statistical Decision Problems
When statistical decision problems of the same type are considered in large groups the minimax solution may not be the “best,” since there may exist solutions which are “asymptotically subminimax.”
...
...