DailyFX Forum - View Single Post - What's in your FSS Portfolio?

Bulldog · #10 (**permalink**) 10-15-2007, 11:40 AM

Some insight into defining your portfolio mix:

Jonathan brings up an important point that each trader needs to consider as they select systems from the application.

I haven’t had a chance to thoroughly review the application yet (I will this week) so I’m not sure what functionality is there but here are a few things traders should consider as they build their portfolios:

Highly Correlated Trades increase the risk in your portfolio.

Traders often look at the Equity Curve of a system or the mix of various systems and tweak and tune exposure until they find an EC and risk metrics that suits them. If they fail to compare the underlying trades in those systems they may be setting themselves up for failure. If the trades in each system have a high correlation, when the market turns around or the longer term price action of the market changes, the models could all collapse at the same time.

Let’s look at the carry trade for a moment; even though there are many technical systems that don’t take into account the interest rate spreads, they benefit from them all the same. I recently finished a consulting project where the base application produced 4 to 5% annual gains in almost every currency pair it was applied to in back testing, except the USD/JPY where it produced nearly 9% returns. The optimization work that had been done up to that point had then been skewed by that bias; when the Yen was removed returns dropped off sharply (in specific configurations) and volatility rose substantially. We saw this phenomenon in almost every times series traded by the application across nearly 100 different variants of the application; even down to the intraday trades.

The reality is it’s very hard to get low correlation in a portfolio and you have to be very careful about how you design your mix: Especially if your models are routed in the same basic theory.

If you’re able to get your hands on the individual trades you can run a correlation on them to see how close they are; you can also look at constructing a classification table of some sort to quickly distinguish how and when the trades are initiated. This is important because almost every trader and model developer (yours truly included) will do his/her best to convince you that their model is the most unique and advanced thing out there… But the reality is, there are only two sides of the market you can be on, and there are a relatively fixed number of events that will trigger price action.

As you apply your classification work you’ll find a large majority of these models are being executed on very similar price action events and are being executed in very similar ways.

It’s no mistake that despite all their advanced technology and mathematics, quant models as a whole took a bath this summer: There is high correlation in the methodology (just like any technical or fundamental based approach) and as a result there was a high correlation in the losses.

A System that Produces 5% a year in the EUR/USD is a good as a system that produces 20% a year using the EUR/USD, GBP/USD, USD/JPY and USD/CHF.

I’ve deconstructed a lot of models that were supposedly capable of 20% a year (or higher) returns. These models have typically had terrible Sharpe Ratios and other risk metrics because of the correlation between the pairs. Unless it’s an outright hedge, there is often very little difference between a long EUR/USD and a long GBP/USD: Yes there is a difference in the spread (transactional impact) and the volatility, but if you look at the impact they have on your portfolio over the long term they will often be very close to the same. Having both of these positions long is typically no different than having two long EUR/USD positions. Why is this important? Because the difference in the two returns isn’t an improvement on actual net pips returned through better trading; it’s only a higher risk profile. Now having said that, the version of my model I’m bringing to the application is also guilty of this… it’s not necessarily a negative thing, but it’s something you should be aware of, especially if you’re going to add leverage to the signals that are sent. In the end you have to know what you’re true risk profile is (not how much money you need to make but how much are you willing to loose) and then be able to break out the real risk level of the model before you add additional risk to it.

I could go on at length about this but I don’t want to bore you to tears on a Monday morning; suffice to say, as you look at models, there will be more that the equity curve that you should consider.

I’ll be providing a lot of detailed information about the model and will be happy to break out whatever I can to help traders make an informed decision. If you have questions don’t hesitate to ask.

John