Antoine Ledoux, Quantitative Research Analyst
December 7, 2022
These days new Machine Learning techniques seem to be coming out every other month claiming to be the next big thing and it is hard to distinguish the good from the bad. So, what better way to compare these models than to have them compete against one another to see which one is indeed the best in time series forecasting (our bread and butter).
This is how the M competitions came about. Brought to life in 1982 by Spyros Makridakis (who gave his name to the event, the Makridakis Competitions, now known as M Competitions), the goal of these competitions is to compare all-time series forecasting methods known to date and see which one performs best. The main conclusion from the first M Competitions (hosted in 1982, 1993 and 2000) was that “statistically sophisticated or complex methods do not necessarily provide more accurate forecasts than simpler ones”.
However, this changed somewhat in the 4th competition (M4) when a data scientist from Uber Technologies surprised the jury with a hybrid approach utilizing both Statistical and Machine Learning methods, beating the benchmark by a whopping 10%. For the first time, the best approaches were hybrid ones: classical statistical models like EWMA or ARIMA were used to create features that were then fed into a more complex Machine Learning model. In M5, the top performers were pure Machine Learning models, beating all statistical benchmarks and their combinations. The latest competition (M6) is still on-going and relevant to our work at Graham, as it will compare forecasting models on price data for 50 S&P500 stocks and 50 international ETFs. May the best forecasting model win!
DISCLOSURE
This presentation includes statements that may constitute forward-looking statements. These statements may be identified by words such as “expects,” “looks forward to,” “anticipates,” “intends,” “plans,” “believes,” “seeks,” “estimates,” “will,” “project” or words of similar meaning. In addition, our representatives may from time to time make oral forward-looking statements. Such statements are based on the current expectations and certain assumptions of Graham Capital Management’s (“GCM”) management, and are, therefore, subject to certain risks and uncertainties. A variety of factors, many of which are beyond GCM’s control, affect the operations, performance, business strategy and results of the accounts that it manages and could cause the actual results, performance or achievements of such accounts to be materially different from any future results, performance or achievements that may be expressed or implied by such forward-looking statements or anticipated on the basis of historical trends.
This document is not a private offering memorandum and does not constitute an offer to sell, nor is it a solicitation of an offer to buy, any security. The views expressed herein are exclusively those of the authors and do not necessarily represent the views of Graham Capital Management. The information contained herein is not intended to provide accounting, legal, or tax advice and should not be relied on for investment decision making.
Tables, charts and commentary contained in this document have been prepared on a best efforts basis by Graham using sources it believes to be reliable although it does not guarantee the accuracy of the information on account of possible errors or omissions in the constituent data or calculations. No part of this document may be divulged to any other person, distributed, resold and/or reproduced without the prior written permission of GCM.
