When attempting to navigate market uncertainty and noise wouldn’t it be great to apply the scientific method to find those elusive signals in the noise? What’s that you say….”this trading is an art requiring the skill of the human participant rather than a conventional science”…..well perhaps this is the choice in the philosophical approach you use to tackle them.
When we backtest over long distance horizons we use the term ‘backtest’ but what we are really doing is identifying the validity of the model we use (our trading system) to describe these markets. We use the backtest, particularly over unseen data, to validate how close that model (a hypothesis framed by our philosophical approach) is in faithfully describing those past conditions. We are applying empiricism as a method to validate our models.
If you don’t use back-tests in your approach to test the validity of your models, then you probably really do not have a model at all. Rather you are elevating your position to an “all knowing’ god, who with this uncanny ability to automatically adapt to the markets gyrations, you are then able to adapt your trading style on the ‘fly’ to capitalise on emerging opportunities. The problem with this stance however is that we cannot validate with any degree of certainty whether such an approach works out over the long term. There just are no available verifiable audited records to substantiate this ‘artistic’ stance.
However we certainly have a verifiable track record of those that do use back-tests as a method to crack open the markets and find that elusive signal in the noise. One of the best of them is David Harding.
The following introduction to Winton Capital is a cracker for those who like to apply a bit of science to their trading philosophy.
In regards to what I mean by “the longer the backtest the better”. This is not a statement regarding an assumption about the future profitability of the model…..but rather whether the model is a sound rationale to use for the future based on it’s correct interpretation of what has happened in the past. It’s the basis for any form of research when planning to launch into uncharted waters.
You may be familiar with academic studies regarding the enduring nature of momentum/trend and in particular ‘absolute momentum’ which is an academic term for trend following. Here is a good example and the following (which is a great read) adds more context to this shady art of trend following and its robustness as a general principle. Well, unlike predictive modelling, price following techniques with very simple rules are used as a basis to make these inferences about the validity of this general principle that all liquid markets trend. All these studies use simple models like a moving average crossover etc. to come to this conclusion.
The further back in time you go in your testing using very simple backtest models, the epiphany suddenly hits you like a brick…..which is what these studies are saying and is central to the idea of the mantra to ‘cut losses short and let profits run’. When we say we need simple trend following systems, the simpler they can get to accurately describe this ‘mantra’ means the more robust they become. What snags the trend follower however in achieving this is a particular variable chosen in their design. They add too many variables for a single chosen system to attempt to squeeze too much juice out of the current condition. Great in principle but no cigar ol’ chap. That’s why you need lot’s of different trend following systems with diversification. One of them is likely to get through the noise without getting snagged. Overall there is a slight edge to trend following….but you gotta bust thru the noise to experience it…hence in addition to the need for many different systems, volatility is used through a breakout technique on entry to bust through the noise.
As you go longer back in time with your backtest, the more headroom you have to see the general nature of the market condition expressing itself. This helicopter view of the long term reduces the random gyrations and noise of the market and more clearly expresses it’s general nature. Also the traditional stepped equity curve of the trend follower over the short term starts to adopt a very linear ascent over the very long term as we zoom out.
If we drilled down by magnifying this view of the backtest, we would see protracted periods of non-trending condition….however from this helicopter viewpoint we see the actual edge of the system we are deploying. You recognise that the edge is very small but persistent….and this is what we would like to ensure before we step off into an uncertain future. This is what we would expect for a generally efficient market with occasional non-efficiencies.
In scaling back and taking a helicopter view on the markets to see what edge you are working with you realise that timeframe does actually matter. Remember that a monthly chart is simply a composite of a W1 chart, a D1 chart, an H4 chart, an H1 chart, a M30 chart….an M5 chart etc. etc. etc. You will not see this fine edge displayed in an M1 chart. There is simply too much noise in it.
If the market was unbounded where there was no end to the lowest timeframe we could go…or the highest timeframe we could go…..then our assumption that the longer term holds more opportunity for trend following would not be rigorous. Fortunately our markets are bounded at the lowest timeframe and open ended on the highest timeframe. At the lowest timeframe we only have a small number of participants interacting continuously. The HFT guys dominate. As we step up in timeframe, a greater population of participants interact including speculators, commercial hedgers, banks and investors. This results in randomness at the lowest timeframes turning into order at the highest timeframes. These interactions tend to cancel out different opinions as we step up to larger timeframes where the winners of this competitive battle are expressed.
Whatever price segment you attack on M5 for example does not infer that it is representative of the longer term edge you see over the 30 year to 50 year timeframe etc. An edge is something that only is revealed by the Law of Large numbers. We might assume that we have an edge trading the M5 timeframe….but over the long term it is a fools fallacy.
The global macro fundamentals like interest rates that are causative factors which drive long term market direction are long term in playing out….not short term. Consumer sentiment and expectation is what is short term and is what predictors focus on. We speculators (that are a very small % of the market) are like piranhas biting on alpha….but have a very small impact on the scheme of general price movement long term. It is only when the markets step outside normal predictability during booms and busts, that all participants including the central banks…take a form of more coordinated action that literally drive the long term direction of price movement. That is when the non predictive trend follower blossoms. The rest of the time….it is a gambling house for speculators or a means to transact at fair value for non-speculators (aka a market).
In referring to this edge and it’s smallness in the scheme of things, we also have another epiphany…..the more I leverage my position sizing, the more ‘wobble’ I am going to deliver to this fine edge that is evident on the long term. This wobble is a speed wobble as you are trying to take on a slim edge too fast with leverage. This is what Ed Thorpe refers to in this video. Leverage is a weapon with a double edge that incorrectly applied kills the trader.
Leverage is not something to be afraid of but we must have an understanding of how to use it. In terms of why you need a long term backtest, leverage is what we do after we extract what edge is available from the low leverage system we use to detect it. We will not be able to see the actual edge in a leveraged system…..you must backtest over the long term to see what edge is available for extraction….before we ramp things up. Hence in undertaking a backtest, the first thing you do is apply the lowest constant leverage you can obtain (eg. 0.01 micro-lots position size) to see the market for what it is. After you have determined the edge you have to work with (which is a really long term measure) you then apply position sizing as a method to gear yourself up in terms that are applicable to the distant past. This way you do not go into an uncertain future with a weak design. A bit like taking a Formula 1 racing car for an off road competition.
We sure don’t know what is happening in the future, but having a suitably scaled robust system to lead you into the future with a model that stands up to the past 20,50, 100, 200 years is the key to the story.
Now let’s assume we don’t know what the future brings but accept the fact that price following and the nature of behavioural markets is the way to tackle them. Also let’s assume that when trading the right edge we need to de-magnify our long term perspective on how markets work….then we may just have a few years of pain while the market mean reverts or is noise ridden and we can’t ride the swells of the long term edge…..well this is why we diversify as opposed to cherry pick which markets to trade. It also reduces our expectations for instant satisfaction and ensures we accept drawdowns as a part of the journey. Fortunately something somewhere is likely to be trending to pay for ‘timing issue’ you are facing on your other markets. This very small edge is happening somewhere some-whence….and under diversification with appropriately scaled systems deployed, we harvest this small edge and through time (like compound interest) lead to wealth building returns.
In a nutshell, while all trend followers agree with the ‘cutting losses short and letting profits run’…you have to go farther than mere belief. Your backtesting should be able to confirm and validate this hypothesis. That’s where the science of backtesting kicks in.
Trade well and prosper
The ATS mob