Friday 16:25–18:00 in Auditorium

Easy Bayesian regularization for fitting financial time series and curves

Dr. Egor Kraev

Audience level:
Intermediate

Description

This talk will show, on a variety of examples, an intuitive yet powerful way of fitting/smoothing financial time series and curves, using a combination of two ideas: firstly, interpreting bid-ask spreads as confidence intervals, and secondly, using Bayesian ideas to easily combine observed values with your a priori assumptions about behavior.

Abstract

When you read 'Bayesian' I bet your first reaction is 'complicated and computationally expensive'; and 'regularization' all too often simply means 'a pesky parameter I have to set'. I hope to convince you they're both easier and more powerful than that.

This talk will show, on a variety of examples, an intuitive yet powerful way of fitting/smoothing financial time series and curves, using a combination of two ideas: firstly, interpreting bid-ask spreads as confidence intervals, and secondly, using Bayesian ideas to easily combine observed values with your a priori assumptions about behavior, such as 'curves must be smooth' or 'skewness is stickier than at-the-money vol' - even if neither skewness nor ATMV are actual model parameters.

Beyond the above, still basically 'linear' approach, the talk will also cover some extensions such as dealing with hard inequality constraints and use of change detection for a responsive nonlinear smoother.

Subscribe to Receive PyData Updates

Tickets

Get Now