20/12/2022, 20:19 Hurst Exponent for Algorithmic Trading - Robot Wealth Menu Hurst Exponent for Algorithmic Trading October 31, 2016 by Kris Longmore Reading Time 3 mins This is the first post in a two-part series about the Hurst Exponent. Tom and I worked on this series together and I drew on some of his previously published work as well as other sources like Quantstart.com. UPDATE 03/01/16: Please note that the Python code below has been updated with a more accurate algorithm for calculating Hurst Exponent. hurst exponent Mean-reverting time series have long been a fruitful playground for quantitative traders. In fact, some of the biggest names in quant trading allegedly made their fortunes exploiting mean reversion of financial time series such as artificially constructed spreads, which are used in pairs trading. Identifying mean reversion is therefore of significant interest to algorithmic traders. This is not as simple as it sounds, in part due to the non-stationary nature of financial data. https://robotwealth.com/demystifying-the-hurst-exponent-part-1/ 1/12 20/12/2022, 20:19 Hurst Exponent for Algorithmic Trading - Robot Wealth We both think that Ernie Chan’s book “Algorithmic Trading: Winning Strategies and Their Rationale”, is one of the better introductions to mean reversion available in the public domain. In the book, Ernie talks about several tools that can be used when testing if a time series is mean reverting. One is the Augmented DickeyFuller test for mean reversion. Ernie also goes into some detail about the Johansen test. Both of these have previously been explored on Robot Wealth and implemented using some simple R code (here and here). Another interesting aspect of testing for mean reversion is the calculation of the Hurst Exponent. The idea behind the Hurst Exponent H is that it can supposedly help us determine whether a time series is a random walk (H ~ 0.5), trending (H > 0.5) or mean reverting (H < 0.5) for a specific period of time. However, if you’ve ever used Hurst, you know that it can be a bit bewildering: not only does it often give unexpected results, but it also returns different results depending on the implementation used in its calculation. Further, there are a few different methods for calculating Hurst; we found that these generally agree for a randomly generated time series, but disagree when we use real data. So how can Hurst Exponent be of value to algo traders? The remainder of this post is devoted to presenting and discussing some Python code for calculating Hurst. In the next post, we are going to delve more deeply into the calculation and work out what’s going on. Our ultimate goal is to demystify the Hurst Exponent and show how to take it beyond some nice theory to something of practical value to algo traders. Without further ado, here is the code for calculating the Hurst Exponent in Python. We determine Hurst by firstly calculating the standard deviation of the difference between a series and its lagged counterpart. We then repeat this calculation for a number of lags and plot the result as a function of the number of lags. If we plot this on a log-log scale, we end up with a straight line, the slope of which provides an estimate for the Hurst exponent. I found this article which describes this approach to calculating Hurst, as does this one. 1. 2. 3. 4. 5. 6. 7. from numpy import * from pylab import plot, show # first, create an arbitrary time series, ts ts = [0] for i in range(1,100000): ts.append(ts[i-1]*1.0 + random.randn()) # calculate standard deviation of differenced series using various lags https://robotwealth.com/demystifying-the-hurst-exponent-part-1/ 2/12 20/12/2022, 20:19 8. 9. 10. 11. 12. 13. 14. 15. Hurst Exponent for Algorithmic Trading - Robot Wealth lags = range(2, 20) tau = [sqrt(std(subtract(ts[lag:], ts[:-lag]))) for lag in lags] # plot on log-log scale plot(log(lags), log(tau)); show() # calculate Hurst as slope of log-log plot m = polyfit(log(lags), log(tau), 1) hurst = m[0]*2.0 print 'hurst = ',hurst You can see in the code that we used lags 2 through 20 for calculating H. These lags are somewhat arbitrary, but based on the best results obtained using synthetic data with known behaviour. Specifically, we found that if we set the maximum number of lags too high, the results became quite inaccurate. These values are the defaults used in some other implementations, such as the standard Hurst function in MATLAB. In Demystifying the Hurst Exponent: Part 2, we will look at these lags in more detail and show how they are actually crucial for calculating Hurst in such a way that is useful and meaningful. We tweak this part of the calculation to uncover a practical application of Hurst in developing algo trading systems. If you have used the Hurst Exponent, or indeed any of the other tests for mean reversion that we mentioned in this post, please share your experiences in the comments. Thanks! Quant trading cointegration, Hurst Exponent, mean reversion, quantitative trading How to Create a Trading Algorithm: So You Want to Build Your Own Algo Trading System? Demystifying the Hurst Exponent – Part 2 15 thoughts on “Hurst Exponent for Algorithmic Trading” Pingback: Quantocracy's Daily Wrap for 10/31/2016 | Quantocracy https://robotwealth.com/demystifying-the-hurst-exponent-part-1/ 3/12 20/12/2022, 20:19 Hurst Exponent for Algorithmic Trading - Robot Wealth Nick Wienholt November 2, 2016 at 10:27 am Hi Kris, I’ve had a brief look at Kaufman’s Efficiency Ratio. The results weren’t great, but I didn’t dig into it effectively enough to fully rule it out as a filter for the mean reversion systems I’m trading. Happy to write it up with a proper analysis. Nick Reply Kris Longmore November 5, 2016 at 10:18 pm Hey Nick I’d love to see a proper analysis of Kaufman’s Efficiency Ratio! You are welcome to share it here! Cheers Reply Nick Wienholt November 9, 2016 at 5:33 pm Thanks Kris – will do – I’ll write it up and submit. Many thanks Reply https://robotwealth.com/demystifying-the-hurst-exponent-part-1/ 4/12 20/12/2022, 20:19 Hurst Exponent for Algorithmic Trading - Robot Wealth David November 5, 2016 at 6:17 am Interesting post thanks… I’m currently investigating applying the Hurst exponent in machine learning to improve my trade selection. Try as I might, I can’t find a default Hurst Matlab function however! Reply Kris Longmore November 5, 2016 at 10:16 pm Hi David, I guess its not really a default function, but try this implementation of the generalized Hurst exponent: https://au.mathworks.com/matlabcentral/fileexchange/30076generalized-hurst-exponent?requestedDomain=www.mathworks.com Reply David November 6, 2016 at 7:15 am Thanks Kris, I’ll give that a try… I also stumbled across this new Matlab function gives 3 separate (!) estimates of the Hurst exponent too: https://uk.mathworks.com/help/wavelet/ref/wfbmesti.html Reply Kris Longmore November 6, 2016 at 3:38 pm https://robotwealth.com/demystifying-the-hurst-exponent-part-1/ 5/12 20/12/2022, 20:19 Hurst Exponent for Algorithmic Trading - Robot Wealth Yes! This is a problem and one that we are going to attempt to get to the bottom of in the next post. Thanks for sharing that implementation. Reply Pingback: Demystifying the Hurst Exponent – Part 2 – Robot Wealth Wei August 8, 2017 at 9:15 pm Hi Kris, Thanks for the article, incredibly interesting. I was just wondering if you could shed some light on where the sqrt of std comes from? I’ve been through both articles you linked to and find the RS method of calculating the Hurst exponent (the one originally derived by Hurst for the work on Nile levels) to be intuitively easier to understand. Looking at your update note, my guess would be that you were initially using that algorithm for calculating Hurst, but then moved on to the Generalized Hurst Exponent calculation that Mike from QuantStart uses on his page? I’ve been through Mike’s explanation, but I still can’t seem to rationalize why we are using the square root of the standard deviation in this algorithm. Any light you could shed on that would be much appreciated. Thanks again for the great article. Reply andy September 8, 2017 at 6:19 pm https://robotwealth.com/demystifying-the-hurst-exponent-part-1/ 6/12 20/12/2022, 20:19 Hurst Exponent for Algorithmic Trading - Robot Wealth Hi Kris, thanks for sharing. I think we can save the sqrt() step here in : tau = [sqrt(std(subtract(ts[lag:], ts[:–lag]))) for lag in lags], and no need to multiply 2 here in : hurst = m[0]*2.0. So we the computation is simplified to: … tau = [std(subtract(ts[lag:], ts[:–lag])) for lag in lags], … hurst = m[0] which saves computation and easier to be understood from the Hurst exponent definition. But still thanks a lot for sharing this, learned a lot. Regards Andy Reply Kris Longmore September 18, 2017 at 9:32 pm Nice one! Thanks Andy, that makes a lot of sense. Reply rando October 6, 2019 at 7:35 pm https://robotwealth.com/demystifying-the-hurst-exponent-part-1/ 7/12 20/12/2022, 20:19 Hurst Exponent for Algorithmic Trading - Robot Wealth Hello, Noob here, why are you taking the square root of the standard deviation of the differenced lags? Thought we were just trying to find the standard deviation? Is it just so they are on a more condensed scale? Reply Curious dude October 7, 2019 at 1:59 am Hello, Why do I get a low Hurst exponent with an only upward moving time series? I get a hurst exponent of almost 0 when using the below generated time series (trends up by about 1). for i in range(1,100000): ts.append(i+1+random.uniform(0,1)) Reply Pingback: Time Series Analysis: Fitting ARIMA/GARCH Predictions Profitable for FX? Leave a Comment https://robotwealth.com/demystifying-the-hurst-exponent-part-1/ 8/12 20/12/2022, 20:19 Hurst Exponent for Algorithmic Trading - Robot Wealth Name * Email * Website Save my name, email, and website in this browser for the next time I comment. POST COMMENT Subscribe to our Newsletter We'll send you insights from the trading desk and updates on quant trading bootcamp. We will never spam you or share your details. Your name Your email Subscribe Most Popular Posts https://robotwealth.com/demystifying-the-hurst-exponent-part-1/ 9/12 20/12/2022, 20:19 Hurst Exponent for Algorithmic Trading - Robot Wealth How to Connect Google Colab to a Local Jupyter Runtime 28.2k views Machine learning for Trading: Adventures in Feature Selection 28k views How to Run Trading Algorithms on Google Cloud Platform in 6 Easy Steps 17.7k views Deep Learning for Trading Part 1: Can it Work? 15.7k views Dual Momentum Investing: A Quant’s Review 15.1k views Hurst Exponent for Algorithmic Trading 15.1k views Time Series Analysis: Fitting ARIMA/GARCH predictions profitable for FX? 13.4k views Demystifying the Hurst Exponent – Part 2 13.1k views Exploring Mean Reversion and Cointegration: Part 2 12.4k views Exploring mean reversion and cointegration with Zorro and R: part 1 12k views Blog Categories Machine learning (16) Deep learning (4) k-means clustering (4) Keras (4) Neural networks (6) Quant trading (94) Backtesting (31) Cointegration (6) Factors (4) FX (16) Options (8) Trading strategies (24) https://robotwealth.com/demystifying-the-hurst-exponent-part-1/ 10/12 20/12/2022, 20:19 Hurst Exponent for Algorithmic Trading - Robot Wealth RW Insider (71) Think like a trader (24) Trading as a business (5) Tools of the trade (77) Python (8) R (50) Time series modelling (5) Trading books (6) Trading infrastructure (8) Zorro (28) Tags activation function algorithmic trading AUD/NZD augemnted Dickey-Fuller test backtesting cointegration data Deep learning dplyr ETFs factors forex fx GBM GPU hedging Hurst Exponent k-means clustering Keras Machine Learning mean reversion options pairs trading profiling profvis put option python quantitative analysis quantitative trading R risk premia simulation slider tidyverse time series trading strategies volatility Zorro microbenchmark momentum opinion selection bias stationarity TensorFlow walk forward analysis Blogroll and useful links Zorro Trader Darwinex Quantocracy The Financial Hacker Ernie Chan’s Quantitative Trading https://robotwealth.com/demystifying-the-hurst-exponent-part-1/ 11/12 20/12/2022, 20:19 Hurst Exponent for Algorithmic Trading - Robot Wealth © 2022 ROBOT WEALTH PTY LTD PRIVACY POLICY | TERMS OF USE | FAQ | CONTACT@ROBOTWEALTH.COM https://robotwealth.com/demystifying-the-hurst-exponent-part-1/ 12/12