We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
The Winner’s curse - Bidding with ML (Niv Geron, Pagaya, PyData TLV Oct 21)
Discover how the Winner's Curse, a statistical phenomenon, affects auction outcomes and learn how Machine Learning models can lead to suboptimal results due to biases and overestimation.
- The Winner’s Curse is a statistical phenomenon that occurs when multiple players in an auction simultaneously correct for biases in their predictions, leading to suboptimal outcomes.
- In machine learning modeling, players use their own models to predict the value of an item, but these models are subject to biases and overestimation.
- The optimal player, who uses a linear regression model with no cubic power, still has a significant bias, demonstrating the difficulty of correcting for biases.
- Features engineered to high powers can lead to distorted results, as seen in the case of the worst player, who uses every feature in its cubic power.
- The Winner’s Curse can occur in various contexts, including auctions with common value and different signals, such as hiring or giving loans.
- It is often difficult to correct for biases, as the optimal player’s strategy may not be applicable to other players.
- The bias is not always the same, and correcting for it in symmetrical auctions may not yield optimal results.
- The distribution of items and the variance of the predictions can affect the Winner’s Curse, making it harder to correct for biases.
- Machine learning models can be used to predict the value of an item, but the Winner’s Curse highlights the limitations of these models in correcting for biases.