Symmetrical vs. Non-Symmetrical System / Robustness

Discussions about the testing and simulation of mechanical trading systems using historical data and other methods. Trading Blox Customers should post Trading Blox specific questions in the Customer Support forum.
Post Reply
Salamander
Senior Member
Senior Member
Posts: 37
Joined: Sun Feb 06, 2005 4:31 pm
Location: Massachusetts/New York City

Symmetrical vs. Non-Symmetrical System / Robustness

Post by Salamander »

Guys: Theoretically, is it supposed to be the case that a system in which your rules for long entry are identical to those for short entry should in the long run be more robust than one in which you tweak the rules a little? For example, I have test results showing a nice improvement in backtesting if I make a tiny adjustment to the short entry rules vs. the long entry rules, but what I have heard over time from more experienced designers is that this will lead to less robustness. I know this is a little vague, but you get my drift. Thanks.
sluggo
Roundtable Knight
Roundtable Knight
Posts: 2987
Joined: Fri Jun 11, 2004 2:50 pm

Post by sluggo »

This thread may be of interest: viewtopic.php?p=11185&highlight=symmetr%2A#11185
damian
Roundtable Knight
Roundtable Knight
Posts: 815
Joined: Tue Apr 15, 2003 8:43 pm
Location: dusseldorf

Post by damian »

As part of your long/short bias testing you should make two portfolios and conduct the repeat tests separately on both portfolios. One of the portfolios should exclude all interest rate products that have carried a long side bias through progressively diminishing inflation. You may find that without the interest rate products included, the benefit of adding bias to the long is reduced. Then look into your crystal ball and see if inflation will now revert to a more typical level, and so too band and note prices (some may argue that there are signs of this being the next big trend over the coming 5 years). If your crystal ball says inflation will return in force, then best to bias the system settings in favour of the short side.

Another issue to contend with: if you trade USDJPY then you will want a different directional biased than if you trade JPYUSD, as per IMM convention. But hang on, long one is identical to short the other. So should you favour long or short trades? Now I'm stumped. Does that mean your long/short ratio bias is a function of quote convention? That causes a problem for swissy as well. What about rate products that can be traded as either price or inversely as yield??
sluggo
Roundtable Knight
Roundtable Knight
Posts: 2987
Joined: Fri Jun 11, 2004 2:50 pm

Post by sluggo »

I did a simple test in which I omitted ALL interest rate futures. You can see which ones were left, in the image file below. Even so, there appeared to be higher performance for Long Only, than for Short Only.
Attachments
The no-interest-rates portfolio
The no-interest-rates portfolio
ips.png (91.05 KiB) Viewed 6080 times
Simulation results, long vs short vs both
Simulation results, long vs short vs both
part1.png (11.42 KiB) Viewed 6076 times
damian
Roundtable Knight
Roundtable Knight
Posts: 815
Joined: Tue Apr 15, 2003 8:43 pm
Location: dusseldorf

Post by damian »

I wouldn't call that a no interest rate portfolio. Besides the fact that you have three bond products, you have the mother of all interest rate products: Fed Funds, with an average winning trade of 22%

I didn't claim that removing interest rate products would make short-only better. I claimed (and have proven to myself) that by removing all interest rate products, the backtest advantage in having a long risk bias is reduced and this is because recent history has seen an unprecedented one way macro trend of falling developed nation interest rates. Unless this trend continues, I do recommend trading a long bias system on a portfolio which is weighted in interest rate instruments.

Using America as a case study, the last 18 months has produces interest rate price action not seen since 1994 when there was a rise in Fed Funds Effective from 3% to 6% in a 14 month period. But even that 14 month run was against the trend of falling rates from 10% to 3% over the preceding 5 years. There are other major periods of falling rates which I have not mentioned, namely the sharp drop from Dec 2000 to mid 2004 (approx 550bps). And the overall macro trend: 1980 FF effective at 19%........ Jan 2004 FF effective at 1%.

If you trade interest rate products then think about giving the short side some risk bias in the next 5 years. Particularly if you trade 90 day euroyen futures.
Attachments
FF.gif
FF.gif (16.59 KiB) Viewed 6038 times
sluggo
Roundtable Knight
Roundtable Knight
Posts: 2987
Joined: Fri Jun 11, 2004 2:50 pm

Post by sluggo »

Boy, that's embarrassing. Although I did manage to remove sixteen bonds and interest rate futures, Damian's absolutely right: I mistakenly left four of them behind.
  • Correctly omitted: Canadian 10 Yr bond (CGB), Swiss Confederation Bond (CON), Euro Bund (EBL), Eurobobl (EBM), Eurodollars (ED), Libor (EM), Euroyen (EY), Euribor (FEI), Long Gilt (FLG), Short Sterling (FSS), Five Year Treas Notes (FV), Japan Govt Bond (JGB), Municipal Bond (MB), Singapore emini Japan Bond (SJB), Australian 10 year bond (YTC), Australian 3 year bond (YTT).
  • Incorrectly included: Fed Funds (FF), Two Year Treasury Notes (TU), Ten Year Treasury Notes (TY), 30 Year Treasury Bonds (US).
Running the simulation again with these four instruments removed, the short-only performance improves just as Damian predicted. The Return Retracement Ratio of short-only went from -0.01 (with the 4 rates), to +0.05 (without them). Trade-Short-only now makes a profit, albeit a modest one (CAGR = +1.7%). Similarly, the Trade-Long-only performance degrades when you remove these four instruments. Return Retracement Ratio drops from 6.57, down to 1.77. Long-Only still beats Short-Only by a comfortable margin (+1.77 vs. +0.05).
damian
Roundtable Knight
Roundtable Knight
Posts: 815
Joined: Tue Apr 15, 2003 8:43 pm
Location: dusseldorf

Post by damian »

One of those "oops" moments... very familiar to me, just ask Tim. Mistake or no mistake, I find this a very tricky topic: structuring system parameters with or without macro market dynamics as an input. I stand by my old opinion: a hybrid system/fundamental trader is the best, but the hardest to be. Systematic expression and execution of strategy, long term fundamental influence on parameter settings (eg, ratio of long-short risk bias). In an attempt to remove discretion, I think it is an interesting idea to dynamically calibrate particular risk appetite functions or bias according to observation of macro, non-price time series. The most obvious example would be CoT series, however it is not quite as fundamental as I am thinking. I have no spare capacity for this line of research at the moment, but it is on the shelf waiting for some work. Another topic.
Post Reply