Discussing the article: "Statistical Arbitrage Through Cointegrated Stocks (Part 8): Rolling Windows Eigenvector Comparison for Portfolio Rebalancing"
Have you ever tried this in live trading conditions?
No offense, but this is just too good to be true!
Have you ever tried this in live trading conditions?
No offense, but this is just too good to be true
Yes, zero offense at all. If it is published it is there to be questioned and discussed. And again, you are right: "this is just too good to be true".
However, please note that I mentioned in the text that I chose carefully the XLK/AAA pair to emphasize the IS/OOS ADF main feature, which is to show past imbalances.
"As said, I cherry-pick this ETF pair to better illustrate the point in question."
Usually this is not as clear as it is in this example, but in this period this specific pair had an evident imbalance that was captured by the test.
Yes, we are using these tests for screening/scoring the baskets and for live trading monitoring. But we are running these experiments within a small account. We are NOT making millions. We are learning what is possible with our limited resources.
Yes, there are opportunities out there, and we know that there are managers using quant strategies for stat arb with different levels of success. It is not easy money. It never was easy money. But in the almost infinite combination of asset classes, timeframes, lookback periods for mean reversion and spread calculation, yes there are opportunities to be hunted.
What I'm publishing here are the methods we are using for finding these opportunities, NOT the opportunities per se. This article is a good example: when we started using the mean-reversion half-time in basket selection we verified the kind of improvement you can see in the two graphs that open the article. This is real, as a method, but these specific graphs are the final result of several choices and MT5 Tester optimizations (this is mentioned in the article too).
I expect people to experiment with the scripts I'm attaching here. The only investment would be your time. I've been using only the symbols included in the Meta Quotes demo account, without Exchange subscription. So, it is free.
There are more than six thousand symbols to be combined and tested in the various timeframes. Also, it is possible to combine them with non-financial data.
There is only one thing we are avoiding as the devil: it is short-term trading. Let alone high-frequency trading (HFT). This in insane for the 'sardines'. :))
I hope this helps.
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
You agree to website policy and terms of use
Check out the new article: Statistical Arbitrage Through Cointegrated Stocks (Part 8): Rolling Windows Eigenvector Comparison for Portfolio Rebalancing.
This article proposes using Rolling Windows Eigenvector Comparison for early imbalance diagnostics and portfolio rebalancing in a mean-reversion statistical arbitrage strategy based on cointegrated stocks. It contrasts this technique with traditional In-Sample/Out-of-Sample ADF validation, showing that eigenvector shifts can signal the need for rebalancing even when IS/OOS ADF still indicates a stationary spread. While the method is intended mainly for live trading monitoring, the article concludes that eigenvector comparison could also be integrated into the scoring system—though its actual contribution to performance remains to be tested.
In this article, we will see how we can overcome the limitations of the IS/OOS ADF validation while extracting the maximum benefit from its strengths. We introduce the Rolling Windows Eigenvector Comparison (RWEC) for measuring the stability of the portfolio weights. The RWEC can perform well with short out-of-sample periods, is less dependent on the in-sample/out-of-sample split point, and can provide us with information about when the cointegration breaks. Getting to know when the cointegration breaks is paramount for live trading monitoring. The RWEC combined with the IS/OOS ADF method should give us a more robust assessment of the stability of the portfolio weights, not only for nice backtests, but also for the system monitoring and portfolio rebalancing on live trading.
We will see how we can take advantage of the combination of methods primarily used to evaluate the stability of the portfolio weights by using them to estimate the stability of the trading signal. We will see how we can rebalance the portfolio when the change in its weights goes beyond the expected, eventually stopping trading the basket if the cointegration breaks. We are going from the scoring system to the live trading signal monitoring.
Author: Jocimar Lopes