The Gateway to Algorithmic and Automated Trading

Feature: Jack Schwager - Making the grade

Published in Automated Trader Magazine Issue 04 January 2007

The number of hedge funds using automated and algorithmic trading has risen exponentially in recent years. That has made it hard for such funds to stand out from each other when pitching to allocators. One such allocator is Jack Schwager, who in addition to an illustrious career on Wall Street and the authorship of numerous investment classics, is also the principal investment manager of the Market Wizards Funds, the flagship fund of hedge funds series of the Fortune Group. AT talked to Mr Schwager about his views on automated trading and some of the criteria he uses when selecting managers.

 Jack Schwager authorship of numerous investment classics, is also the principal investment manager of the Market Wizards Funds, the flagship fund of hedge funds series of the Fortune Group.

Jack Schwager

Are you noticing an increase in the number of automated trading programmes?

At Fortune we haven't noticed an enormous increase, but then autotrading has been around for quite a while. It certainly predates the popularity of hedge funds and was first seen in the CTA community.

Perhaps that lack of dramatic growth has something to do with the competitive nature of automated trading - especially in areas such as statistical arbitrage. Doing it well is demanding and typically requires a very sophisticated operations function.

What are you looking for when you examine an automated trading fund?

Often we are approached by managers who only have a simulated track record; not a real one. It is theoretically possible to do simulations that approach real results, but there are numerous challenges. The first problem is that most people don't follow the appropriate rigour. Secondly, even if they did you have no way of knowing for sure. Thirdly, a lot of people tend to fool themselves when it comes to simulations. Even if each test is in itself statistically valid, if you do enough tests then ultimately just through sheer luck you will hit something that looks good. The numbers may appear sound, and as a stand-alone statistical process be valid, but because they are the product of an overall research process that has rejected lots of other approaches they are a lot less statistically reliable than you might think.

Therefore I tend to rather throw up my hands when presented with simulated results. Even those who have conducted their simulation in a rigorous manner will unfortunately find themselves lumped in with those who haven't, simply because there is so much misleading simulated information out there. As a result, the first question I always ask is whether the results are simulated or real.

Showstopper: Lack of a real time track record

And if they are real?

If the results are based on real activity, then one moves on to consider other questions. How long has the programme been running? How large a trade sample does it have? How statistically reliable does it appear? If you have something that has traded frequently in real time then there is potentially some statistical reliability.

As regards the trading rule set, you're never going to see the exact model that is being used. From a proprietary perspective that is entirely understandable and I don't even expect it. However you do want to get some sort of general sense of the approach being used. Depending upon what the approach is, it may be one that has generally lost its efficacy. If it is, you want to be looking for some sort of reasonable explanation as to why the manager has been able to make it work where others haven't.

What about the manager's risk management process?

You obviously need to look at what the risk control process is and whether it is consistent with the trading approach. For example, if somebody is using a trend following approach (which is fairly common) this has the problem of being relatively volatile in terms of returns. However, it also has some sort of intrinsic logic around risk control because if the position goes against the manager then they will by definition be getting out of it, or at least reversing.

By contrast, if it is a counter trend approach there is the inherent problem that the more the market moves against the manager, the stronger their entry signal would become. Therefore a manager using a counter trend model needs to have a robust explanation of how they deal with the quandary of a methodology that by its very nature runs counter to the concept of risk control.

High-frequency automated trading has obviously become increasingly popular in recent years. Do you regard that as more or less risky than lower frequency approaches?

High-frequency trading is intrinsically much less risky because of what happens when things go wrong. The more traditional approaches, such as trend following, tend to be longer term in nature. The problem with that is by the time you accept that you have gone wrong on a trend a large loss has occurred - thus causing volatile returns.

To some extent I think high-frequency trading has grown out of the realisation that these longer term approaches are inherently volatile. The way to get around that is to limit each trade in terms of its time span and the magnitude of its permissible movement against you.

Therefore, when you do many such short-term tightly stopped trades, each individual trade will have very little impact - so if things go wrong they will at least go wrong in slow motion.

For example, assume you have a high-frequency, lowvolatility trading model that was averaging a return of 1.5% per month but then stops working. Your risk as an investor is that this approach had some consistency but now it isn't providing any return. So after three, six or nine months of flat performance without a good explanation you will simply get out, but you haven't actually lost a lot. Compare that with a longer term trend following model that gets hit by major reversals - there you may be looking at possibly a 30-40% loss on your capital.

"'re never going to see the
exact model that is being used."

But surely, in view of the latency implications, an automated high-frequency trading approach runs the risk of losing its edge for technological rather than pure performance reasons?

Yes, it's certainly true that some of the edge from such a model derives from efficient split second order execution. As more competition arrives, that edge is arbitraged away. However, I would reiterate my earlier point that you can clearly see this in the numbers.

Therefore your risk is limited because the decay is probably quite gradual. There is also an offset to this, which is that costs keep on declining - both for the technology to support the methodology and for trading fees. That helps to counterbalance increasing competition at the highfrequency end of the spectrum.

Apart from latency, surely the number of competitors looking for an edge in the highfrequency spectrum means that their trading models' performance must decay more rapidly?

As an investor, that isn't really your problem - it's the manager's. From your perspective (assuming there aren't any other extraneous problems) you are unlikely to lose too much money on a high-frequency programme. The other advantage is that because they are trading so frequently you can quickly make a statistical determination because the high trade count rapidly builds a statistically significant sample. By contrast, with a conventional trend follower you might have to wait twenty years before you gather enough data! As a result, I hardly ever invest in trend followers because of what I see as the unacceptably high volatility of returns and the length of time you would have to wait to make the decision that the manager's approach has lost its efficacy.

So once you have invested in a particular high-frequency trading programme, how long before you know it was the right choice?

I think you have to give it six months minimum - otherwise you're not being fair to yourself or the manager. After six months you should have a large enough trade sample to decide whether to quit, hold, or scale back your investment. The exception to that is if one month is wildly out of line; then of course you can make an immediate decision to quit.

Do you find it attractive when people leverage automation as a means of increasing diversification through additional models/timeframes/markets?

I'm a big fan of diversification in any guise - it almost always has a beneficial effect. (Unless perhaps one is adding inferior inputs in order to obtain diversification.) More diversification will improve the return to risk, which is my primary performance measure.

Return alone is a meaningless number; it's rather like an internet site quoting international hotel rates without specifying the currency. The currency denomination of return, so to speak, is risk.

Do you have any preference for the type of diversification (e.g. by timeframe or model)? And have you noticed any particular trend among managers in their preference for type of diversification?

I don't think timeframe and model diversification are mutually exclusive - both are beneficial. Timeframe diversification at the shorter end of the spectrum is largely a matter of technology once the technology to handle the tick data in very large volumes at high throughput rates is available. Then there is theoretically no reason you shouldn't take the same model and apply it to shorter timeframes. You can't hurt the return to risk by doing that, you can only enhance it, which is perhaps why it is becoming more common.

With regards to managers' diversification preferences, I would say more managers are claiming to diversify by model than timeframe at present, but both are pretty commonplace.

Have you noticed any discernible recent trend in terms of the type of methodology being used by automated trading programmes?

Nothing hugely significant - though we are perhaps seeing slightly more emphasis on counter trend models than we used to.

How many managers a month are you trying to evaluate?

It's extremely hard to say. All sorts of possibilities cross my desk - some are screened very quickly and others will be followed through more intently. Over the course of the year I think it's probably safe to say we are looking at several hundred.

"I don't think timeframe and model
diversification are mutually exclusive..."

When vetting possible managers are there any major turnoffs?

As I mentioned earlier, insufficient real time track record. Another area we scrutinise closely is where/when is the money being made and where/when it is being lost. Some approaches appear superficially adequate, but when you look more closely it emerges that the bulk of the money is being made in good times, but when the market has difficult months the manager loses significantly.

That type of manager is only going to exacerbate the volatility of a diversified portfolio. The overall returns might look good, but when they lose money they will be doing so at exactly the wrong time. So in brief, managers whose performance is closely correlated with the market they trade or hedge fund indexes are unattractive.

Finally, a murky approach to risk control is an automatic negative as is obviously any type of dishonesty.