Managing Correlated Trades with Covariance and Stress Awareness
Summary
The document recommends using a covariance matrix to identify whether a portfolio of trades is concentrated in the same factors or bets. The matrix can be estimated from a factor model or from the prediction signals produced by multiple strategies, giving an ex-ante view of historical relationships among trades. Designing the matrix requires judgment, and the answer points to estimation research as relevant background.
A second answer cautions that historical correlation is unstable. Correlations can rise or shift during market stress, so a portfolio that appeared diversified in calmer periods may behave differently in a crisis. The discussion cites several past stress episodes as examples and lists research papers on improving covariance and correlation estimation. It provides no specific monitoring algorithm, trading results, or preferred estimator. The practical lesson is to treat covariance analysis as an input to exposure management while recognizing estimation error and regime changes as important limits.
Key ideas
- A covariance matrix can reveal trade concentration in shared factors or model signals.
- Factor models and prediction signals are two possible bases for estimating trade covariance.
- Historical correlations can change, particularly during periods of market stress.
- Covariance estimation involves judgment and can contain substantial error.
- Diversification based on calm-period correlations may not hold in stressed markets.
Tags
Full text
# Minimizing Correlation # Minimizing Correlation Is there a quantitative method in monitoring trades to reduce the possibility of correlated trades? ## Answer by Ram Ahluwalia (score 5, accepted) https://quant.stackexchange.com/a/1144 The most straightforward approach is to develop a covariance matrix to ensure that you are not overweighting to the same factors or bets in your trading. The covariance matrix can be built off of a factor model, for example, or you can construct a covariance matrix based on your prediction signals if you have multiple models. In this way you can understand the ex-ante historical correlation of your trades. Note that there is a considerable amount of art in designing a covariance matrix (See Fabozzi). Without understanding more about your approach it's hard to be more helpful than the above approach. ## Answer by bill_080 (score 3) https://quant.stackexchange.com/a/1145 Historical correlation isn't as useful as you might think. Like volatility, correlation is not constant. During times of stress, it is common for the correlation of many different assets to increase/change. As examples, look at the data for August 2007, the last quarter of 2008, and May 6, 2010. Any trading scheme that minimized correlation before those periods probably had a much different affect during those periods. Edit 1 (05/10/2011) =========================== I've bumped into all sorts of problems with this issue, with the estimation itself involving large errors. If you dig around, you'll find several papers with important improvements. http://www.ledoit.net/honey.pdf http://arxiv.org/PS_cache/arxiv/pdf/1009/1009.5331v1.pdf http://www.oxford-man.ox.ac.uk/documents/papers/2011OMI08_Sheppard.pdf http://www.christoffersen.ca/CHRISTOP/2007/RM2006_introduction.pdf http://www.kevinsheppard.com/images/4/47/Chapter8.pdf http://www.kevinsheppard.com/images/d/de/CES_JFeC.pdf http://www.kevinsheppard.com/images/c/c6/Patton_Sheppard.pdf
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.