Page 1 of 3

fighting the randomness: healthy system vs. tradable system

Posted: Mon Jun 19, 2006 2:42 pm
by michal.kreslik
Healthy system vs. tradable system

there is a substantial difference between the healthy system and a tradable system:
  • healthy (statistically & mathematically sound) system is able to uncover some price behavior inefficiency and generate trades that deliver consistent non-zero outcome with no trade costs involved
  • tradable system is a combination of a healthy system and a specific market (specific trade costs, volatility, liquidity) which delivers a positive outcome
Now this might seem plain, but please think about it a littlebit.

If you come up with a healthy system that uncovers some market inefficiency (like buying the exact bottoms and selling the exact tops :D ), then it's splendid. But now imagine the market you will trade this system on will be a non-volatile, non-liquid market with tiny swings and feeble vitality in general.

Yes, you will be able to catch all the swings perfectly, so the mathematics behind your idea works correctly and the little nerdy scientist inside you claps his hands =D>

But given the fact that the average swing would be too minute to cover the spread/commissions (to say nothing of the ugly slippage in stocks if the market is not liquid enough), the hard-headed investment manager inside you sets in and informs the little nerdy scientist that the system is an overall loser. (well actually, there might be a number of persons inside you expressing their opinions on the issue. If this is the case, I recommend go and see your shrink)


Healthy is not always tradable

What's the morals of the above example?
  • You might have a healthy system (does it catch all the swings correctly? Yes)
  • This healthy system is not necessarily a tradable system when combined with the particular market (is the outcome of applying this healthy system to a specific market with all the costs involved positive? No)
Don't hang your head. Your pet healthy SwingSystem 01, as we are going to call it, may still work profitably on another market where trading costs are lower (lower spread, higher liquidity/lower slippage) or the volatility is higher, or both (preferably 8) ).


Fighting the randomness

Now let's focus on my initial definition of the healthy system:
  • healthy (statistically & mathematically sound) system is able to uncover some price behavior inefficiency and generate trades that deliver consistent non-zero outcome with no trade costs involved
Let's have a closer look at the "non-zero" and "no trade costs involved" factors.

If you come up with a trading system and its equity curve goes nose down with the trading costs involved, you actually can't assess whether it is a healthy system or not.

It might still be a healthy system.

Why? :smt104

Because the trading costs distort the assessment at this stage of development. Remember, the trading costs apply to a specific market only and have nothing to do with the trading model logic itself.

If the trading model generates consistent non-zero outcome without the trading costs involved, then it doesn't matter whether the outcome is positive or negative. You may easily switch between the negative and positive outcome by simply reversing the trade logic.

But if there are trading costs involved, you don't know wheter the nose-down equity curve is a consequence of a desirable healthy model that generates consistent non-zero (in this case negative) outcome or a miserable not healthy model that generates random trades (random = zero in the long run) and the non-zero behaviour is only a result of subtracting the trade costs.


A textbook example

To illustrate, let's have two trading systems:
  • SwingSystem 01 (a healthy system of yours)
  • SwingSystem 02 (a not healthy system of your neighbor :lol: )
two markets:
  • Market A (trade costs = $10 a trade)
  • Market B (trade costs = $60 a trade)
and two directions:
  • direction 1 (go long when long signalled, go short when short signalled)
  • direction -1 (go short when long signalled, go long when short signalled)
This leaves us with 8 (2 to the power of 3) possible system/market/direction combinations to explore.

The healthy SwingSystem 01 is able to deliver a consistent non-zero outcome with no costs involved and it does not matter whether the outcome is positive or negative as long as it is non-zero (non-random):




On the other hand, our poor neighbor's not healthy SwingSystem 02 generates random trades (zero outcome in the long run) and he can't make it better even if he switches the direction from 1 to -1 since randomness (or zero) is not polarized, there is no directional trend (it always ends up yielding zero):




Now let's apply our good, healthy SwingSystem to Market A and Market B with direction 1:


Obviously, the healthy SwingSystem 01 lives up to the expectations to make up for a tradable system only at the Market A, where the trading costs are $10 per trade. Market B features $60 trading costs per trade, which makes the combination of this healthy system and this market untradable.

We now know our SwingSystem 01 produces a positive tradable outcome for a direction = 1 (above left), so, naturally, reversing the trade logic to direction = -1 would make no sense for SwingSystem 01:





Now whatever the Market and whatever the direction, the not healthy system always produces a negative outcome. This non-zero outcome is caused by an inclusion of the trading costs (always a negative tendency), not by a desired ability of the system to produce a non-zero outcome. This is the prinicipal reason why the assessment whether the system is healthy or not cannot be done with the trading costs involved. Random SwingSystem 02 with trading costs accross two markets and the opposite directions:





Conclusion

The only way to capitalize on price behavior inefficiency is to find a set of rules that produces a series of non-random observations. It's irrelevant whether the observations (the trades) have a negative tendency or a positive tendency.

Since including the trading costs in a trading outcomes calculations always adds a negative tendency to the original string of non-biased observations, it is essential to not include the trading costs in this stage of trading system development.

If this rule is neglected, the trading system developer is not able to tell whether the non-random behavior with non-zero outcome is caused by a sound set of rules or simply by a negative tendency that has been falsely implanted into the series of observations by adding the trading costs.


Have a very nice day, friends!
Michal

Attached: the source excel table with graph

Posted: Mon Jun 19, 2006 4:16 pm
by eudamonia
Michal,

Several good points here and a great presentation.

I believe that examining systems the way you describe above is crucial in determining whether you have identified patterns that are non-random. Also, I would add that later in strat development when one has determined the health of the model, one should definately consider execution costs as part of the market itself. The reason for this is that your "healthy strat" may work well with no execution costs on 1 minute bars taking 1000 trades a month, however, no strategy in the world can survive those execution costs in a real market. You may need to adjust your strat to work on 30 min bars and only take 100 trades per year.

The point of the above statement is that some healthy strategies will never become tradeable and some can be modified to become tradable.

Edward

Posted: Mon Jun 19, 2006 4:31 pm
by michal.kreslik
Sure, Ed,

a "market" in this respect is determined by not only an exchange/symbol specification, but by the optimal timeframe, too.

The higher timeframes of the same symbol always offer more volatility and lower relative transaction costs, although the volatility decreases with the second power in relation to the time.

I did pretty interesting research on the theme of ranges versus timeframes and their mutual relationship, I'll post it here some day.

Michal

Posted: Mon Jun 19, 2006 4:39 pm
by eudamonia
Michal,

That sounds interesting. I'd definately like to see that.

Edward

Posted: Mon Jun 19, 2006 10:36 pm
by jhtumblin
I think that it's a no-brainer when developing a system to take into account your overhead (commissions, platform fees, maintenence fees, etc). Anytime I come up with something I think is good I won't even backtest it until I have my commissions and slippages in place. This way I don't get that "false hope" that some may achieve when they see the profit on top.

The only healthy systems are tradeable ones...the bubble boy never got any real experiences until he made it outside his universe. A weak analogy but nevertheless you get my point.

Also note that for some people, (myself not included) huge drawdowns are not that big a deal. Afterall if you can sit through a 50k loss just to reap a 50k gain, who with a large amount of capital wouldn't be interested?

Posted: Mon Jun 19, 2006 10:59 pm
by eudamonia
jhtumblin,

Historical drawdowns are not necessarily a worst case scenario. The markets a driven by non-linear statistics producing fat-tails. What that means is that your worst case scenario could be a whole lot worse than your historical testing suggests. Sometimes by a factor of 10 or more.

I'd say that a 50K gain vs a 500K MaxDD would look bad in almost anyone's book.

Posted: Mon Jun 19, 2006 11:00 pm
by michal.kreslik
jh,

it would be a wasting if you rejected a healthy system just because it performs poorly with a particular market/timeframe combination.

Backtest, Commission, Slippage, Money Management, Scaling

Posted: Tue Jun 20, 2006 12:21 am
by heyen
It would be horrible to waste a good system because of wrong approaches.

A healthy system can be improved greately by using position scaling and money management.

Many stable systems are profitable with a winning ratio below 40%.

Posted: Tue Jun 20, 2006 3:59 am
by jhtumblin
I think you guys are reading into my posts too much... Too smart for your own good 8)

You assume I am not applying money management or attempting different timeframe combinations to my strategies, and I assumed all of that was implied. I guess not.

Either way I see your points but my point is that for a system to be tradeable yet healthy at the same time you need a couple of things:

(I am taking a technical, non-fundamental approach to the market with these ideas)

1) A consistent formula that arrives at appropriate or near correct answers taking into account most common variables. ( by appropriate I mean gets you at least trading in the right direction initially)

2) Factors in your trading overhead so that at a minimum, your costs are covered.

3) Has a minimal risk factor (money management).

The problem we face when weighing whether a system is healthy derives from the fact that not everyone's idea of these points is the same. For instance in regards to point 1, scalpers may consider the right directions for 8 pips is tolerable whereas swing traders are looking to capture several hundred pips.

Another example is right above in Euda's post. He may not think that waiting 2 weeks through a huge DD for profit is viable, whereas others may see it as a perfect method of trading.

If you want a view of my ideal trade model and my personal requirements, I have no problem sharing that, however I can guarantee they will not match everyone else's.

In my opinion, profits are profits, we all search for a holy grail of course but one man's holy grail is just a wooden cup to someone else.

Healthy vs tradable

Posted: Tue Jun 20, 2006 8:21 am
by heyen
jhtumblin,

please, dont misunderstand me.

I wasted weeks to find wrong backtest settings. And i wondered why even the most classic approaches failed.

Im dead mad at myself because i didnt keep the many good approaches i have to remodel now. And sure if its only more risk than gain after commission and slippage its no good at all. I dont make up the perfect system out of the blue. For me its hard work and many remodelling phases.
I like to have the trades marked in the historic data and then go through them to see what didnt work. As someone said:
"It is easy to see the trades that would have worked, but you dont see the trades that wouldn't."

So often it is some simple modifications of the strategy that turn my "loosers" into "rockets".

And sure, share your holy grail with us. I like to think throuh others ideas as i get stuck very often in my systematic approaches.

My approaches are in this section of this forum.

Thomas