Why You Can’t Predict the Market.

Past Performance is no predicator for Future Performance.
The infamous line plastered across the bottom of most if not all investment products you see nowadays.
It’s there for very good reason as well.

As demonstrated by the financial crisis the market is full of unknowns and actually some system or strategy that works in the past may not work in the future. Yet each day millions are spent trying to predict the future and my team and I are a tiny part of this group.
The reason for doing this is somewhat obvious, the payoffs if right are extremely large; however the risk and success rate is tiny. Afterall you only hear about the people that make it big or spectacularly fail, you never really hear about the majority who simply don’t succeed.

Ok so let’s break down the challenge.

1) The first assumption many of us have to make is that the efficient market hypothesis does not hold true

Now for many of you purists out there this is an initial leap too far.
But in order to spot patterns in the market you need to believe fundamentally that there is a chance that the pattern will repeat itself and that you can base some future knowledge on a current or past state. This by its very nature however is a very difficult concept to prove or disprove. You can prove that the purist most stringent form of the efficient market hypothesis does not hold true; however semi strong and weak forms are a lot harder.
Ok but let’s assume we have made that leap and by proxy of people outperforming the market there must be some reason to that.

2) We then need to consider the various potential variables for spotting patterns and how these conditions may play out. More critically we need to attempt to work out when a series of X inputs results in y behaviour

Here we get a big data problem that is almost unfathomable to understand in its purest form.
If we consider just the most basic concept, that past price gives some indication to future price.
There are thousands of ways to manipulate a single data point such as the current Bid rate in a single currency that mathematically spotting a single pattern that even has a high probability pay-out is extremely difficult to do and the word extremely is probably a long way off an understatement.

In our systems alone when building a single portfolio system we have a search space of over 3 million components, each component consistent of a single currency pair with a single set of a parameters. We aren’t even close to scratching the surface on a universe of possibilities and yet for us to consider a single portfolio for say 200 strategies we need to employ specially designed algorithms just to ensure that the search time is cut down to something reasonable (our life time is a good start, a week even better).
Our search space is this instance is 200 strategies (k) from 3 million components (n).

Which gives us the potential for a number I don’t wish to calculate it’s so big…

Out of this search space we are looking for the single portfolio that represents the most efficient and perfectly regular set of
parameters that allow us to find a pattern in the markets.
Where this becomes scary is that this search space is a tiny amount of how big we could make this and we only include two very basic ways of manipulating one very small parameter.

Imagine if we took even the simplest technical indicator that may have 2 or 3 parameters as well as multiple conditions. What is the probability of you finding a particular pattern and aligning it to specific conditions in the market?
Therefore my first statement is probabistically correct, even if there are millions of people reading this and they all worked in financial institutions and could be considered (sophisticated investors), the chances of them actually stumbling across a specific pattern that works on a consistent basic is almost 0.

3) But, but, but

Ok stupid title but this is where we get into the real semantics that are critical to all of these arguments.

See you don’t need a full proof method for predicting the market. Even if we assume there are millions of them in the infinite search space the chances of any human ever finding them with today’s technology is still zero.
But we don’t need to.

What we need to do is prove that patterns are statistically significant for specific conditions.
So what do I mean?

Well the key here hidden in my previous comments that we need to find conditions and probabilities to that given current conditions one of several potential patterns may play out with different probabilities and each pattern has a different payoff.
This then becomes a risk calculation whereby like any company you take all potential trades whereby the expected payoff outweighs your hurdle rate.

And here lies the big data problem.

How can you ever work out the chances or probability of a system, strategy or particular pattern and its potential various payouts?
Well actually that is easier than it looks if you have well defined trading rules.
Simple back testing, paper trading can give you some good clues to your trading strategy all you need is a decent sample size covering multiple trading conditions and you will have a good understanding of the potential success rate.
Once you have this success rate you can work to tweak the risk management model to ensure you minimise your downside and maximise your profit potential.

When you do this, it turns out you don’t need to predict the market, actually there are countless strategies out there that if you stick to the rules and are consistent you can simply follow them and earn a decent consistent return.
So my advice is stop trying to do the impossible and stop looking for that perfect system, start focusing on what you can trade, your risk and money management and ensuring your consistency.
Translate to English Show original