Search
× Search
Friday, December 20, 2024

Archived Discussions

Recent member discussions

The Algorithmic Traders' Association prides itself on providing a forum for the publication and dissemination of its members' white papers, research, reflections, works in progress, and other contributions. Please Note that archive searches and some of our members' publications are reserved for members only, so please log in or sign up to gain the most from our members' contributions.

Pair trading: ADF, Hurst, Variance Ratio Test using daily or 5 minute data is better?

photo

 Chang Min (Leo) Chu, Quantitative Research Analyst at Antilles Capital LLC

 Wednesday, May 28, 2014

Hello, I have used 5 minute and daily data to compute the above test. For 5minute data, it shows very strong mean-reversal but daily data suggest otherwise. They are using the same period of 3 months data. Does that means I should use 5minute data to trade and fit the models?


Print

4 comments on article "Pair trading: ADF, Hurst, Variance Ratio Test using daily or 5 minute data is better?"

photo

 private private,

 Thursday, May 29, 2014



Depends on the hold time you're targeting. If you're targeting smaller hold times (intra-day or once in a few days), you should use the higher frequency data. However, the fact that you're seeing non-stationary behavior in daily data implies that the pair is in a trend over longer periods and you could end up losing a fair bit if you're invested during that trend. So you need to make sure that your strategy stops out during those trends, and that the loss is more than covered by your profits during the mean-reversion phase.

Another important attribute that can bite is that the variance you see using 5-minute data may be too small to cover the bid-ask spread. Using longer periods will give you more variance and the percentage you lose on slippage will be a lot lower, albeit at the cost of longer hold times.


photo

 Dr. Adam Ginensky, Lecturer at MSc in Analytics at University of Chicago Graham School

 Thursday, May 29, 2014



Assuming the point is to make money, the only sensible answer is for you to backtest your trading strategy (over varying time periods). As Ravi points out, you better include bid/ask spreads and slippage in your calculation.


photo

 Chang Min (Leo) Chu, Quantitative Research Analyst at Antilles Capital LLC

 Friday, May 30, 2014



Thank you Ravi and Dr Adam


photo

 Bharath Rao, Entrepreneur

 Tuesday, June 3, 2014



Chang,

1) 3 months is too small

2) A time series can have trending and mean reverting behavior at the same time but over different time horizons.

3) Have you established any frequency for mean reversion?

Lastly, feel free to fit any model you like for the 3 months data you have. But validate the model against data that has not been used in arriving at the model.

Please login or register to post comments.

TRADING FUTURES AND OPTIONS INVOLVES SUBSTANTIAL RISK OF LOSS AND IS NOT SUITABLE FOR ALL INVESTORS
Terms Of UsePrivacy StatementCopyright 2018 Algorithmic Traders Association