Podcast: Gordon Ritter on optimal execution and reinforcement learning

How often should an investor trade? It’s a simple question, but answering it can be fiendishly complicated. Quants have proposed several solutions over the years, starting with Grinold and Kahn in 1989. Their rule, known as the fundamental law of active management, holds that a manager’s success depends on the frequency with which they trade.

But that ignores transaction costs, which can erode the returns of even the most well-conceived strategies. In 2001, Almgren and Chriss introduced a solution that takes transaction costs into account, linking optimal execution to trading volumes and volatility.

More recently, in 2016, Garleanu and Pedersen presented a closed-form optimal dynamic portfolio strategy with predictable mean-reverting returns and transaction costs.

In this episode of Quantcast, Gordon Ritter, chief investment officer and founder of New York-based hedge fund Ritter Alpha, describes an even simpler solution, which he co-developed with his colleague Jerome Benveniste, senior research scientist and co-head of equity statistical arbitrage at Ritter Alpha, and Bastien Baldacci, an adviser for the equity derivatives team at HSBC Paris.

“Our main result in the paper is a direct generalisation of Garleanu and Pedersen,” says Ritter.

The formula links the optimal turnover, which is defined as a fraction of the overall book size, to the autocorrelation of trading signals, the investor’s risk aversion and the liquidity and volatility of the asset. One of the advantages of this approach is that it uses variables that traders are familiar with. Its simplicity also means it can be used at any frequency. Ritter says that, at his fund, the formula “is updated more than once per hour”.

The limiting feature of the formula in its current form is that it is only designed for single-asset cases. And while the approach can be extended to a portfolio of independent assets, the more realistic case of a portfolio of correlated assets is yet to be addressed. The main challenge is incorporating liquidity and correlation matrices and managing the higher dimensionality this introduces. “That is something we are working on, but haven’t finished yet,” says Ritter.

As a quant hedge fund manager, optimal execution has been a central element of Ritter’s trading framework for a long time. He has published several papers on the topic. His 2018 paper, Machine learning for trading, showed how a multi-period trading problem can be solved using reinforcement learning (RL). That work, for which he was named Risk.net’s buy-side quant of the year, opened the way to more research into several RL applications ranging from trading strategies to robo-advisory.

Ritter discussed the state of RL in finance and explains how it can potentially be applied to any problem that has a time-dependant feature.

Index

00:00 Introduction

02:50 The trade execution formula

08:00 Turnover and alpha correlation

10:55 The Gaussian process and linear price impact assumptions.

14:15 How the formula is used in practice

18:08 Single-asset and multi-asset cases

24:55 Reinforcement learning and the applicability and evolution of reward functions

31:45 What can RL help with

37:25 Current and future research projects

To hear the full interview, listen in the player above, or download. Future podcasts in our Quantcast series will be uploaded to Risk.net. You can also visit the main page here to access all tracks, or go to the iTunes store or Google Podcasts to listen and subscribe.

Now also available on Spotify.