Is 100 percent mechanical trading possible?

[corei]

I am yearning for a 100 % mechanical trading system that is profitable?. Is this possible?

 

[donaldduke]

I am yearning for a 100 % mechanical trading system that is profitable?. Is this possible?

They are possible but can you trade them?

Typically a mechanical system will have a 2:1 or less yearly profit/drawdown ratio.

This means if you want to make 100% a year, you have to accept 50% drawdowns. Or if you want 50% a year you have to accept 25% drawdowns.

Also there will be a distribution in returns, if the average you are aiming for is 50%, then one year you might make 100% and the next 0%. Do you have the discipline to trade a system everyday for a year and not make any money that year to show for your efforts?

Most mechanical systems are right less than 50% of the time, most in the 35% to 45% range, this means you will suffer long losing streaks at times. Maybe 10 losers in row, sometimes you might only get 2 small winners in 20 trades.

You need to do lots of backtesting and have lots of discipline and you cant second guess the system either. You have to take every trade, you never know which ones will be the big winners.

 

[MrGecko]

Apologies for a silly response, but it really does depend on your strategy, and by how you determine "mechanical'.

If by mechanical you are referring to a series of conditions based on technical indicators (such as a very simple buy on oversold RSI when Last price > moving average A > moving average B) then I think you will find yourself out of luck.

However, that does not mean to say that there isn't a way to program a profitable strategy into a computer, and all you have to to is the childminding - it just means your strategy has to be a little (a lot) more robust.

For example, wasn;t one of reneissances' strategies to watch for action in the put markets, then when there were large volumes going through they bought the underlying stock and held for 3 mins (?? or something like that, front running the desk that was buying the insurance before the risk asset). It is wholly feasible to let a computer do this on it's own, or stat arb, convertible arb, whatever... but you need sh!t hot technology and double smart b@stards just in case.

So yes, I will say it is possible for systems to run without human intervention, just to be there in case (what is the joke? trading desk in 10 years will have a computer, a dog, and a trader. It's the computers job to trade, the traders job to answer the phone and feed the dog, and the dog's job to bite the trader if he goes anywhere near the computer). But I will wager that the systems that are 100% mechanical are a little more grounded than the moving average example outlined above.


End of the day: Yes, but the barriers to entry are massive, in terms of capital, talent, IT hardware, and so on...

 

[nine]

Yes.

But.

As pointed out in the above posts, mechanical systems tend to underperform good discretionary trading. I didn't really realize why until I recently started to mechanize elements of my rules. These rules are written down and I (thought) follow them rigorously on a good day.

But when you mechanize them you discover how much discretion is involved.

Also, more importantly, the discretion is driven by context. And the context is very hard to program. How do you say "you take the trade if its early afternoon, has tested a significant S&R cleanly and ...." to your computer. What most seem to do is build systems that don't do that and as a result you have higher drawdowns ... but in return you don't have to sit and filter for context before every entry.

 

[nine]

Yes. I love the new avatar. An extra gold star for that. Actually, time I gave some other people stars ... been working too damned hard


Do you like my lurker in the stochastic darkness?
They tend to take out the odd Tourist in this part of the world but they are very patient predators - only the best tourists will do - something for traders to aspire to always.

 

[MrGecko]

avatar - fine. Above average, but not spectacular.

lack of posts? unsatisfactory. There was a time, nine (mod status aside) when you spoke and people listened...

 

[Citizen2007]

Yes.

But.

As pointed out in the above posts, mechanical systems tend to underperform good discretionary trading. I didn't really realize why until I recently started to mechanize elements of my rules. These rules are written down and I (thought) follow them rigorously on a good day.

But when you mechanize them you discover how much discretion is involved.

Also, more importantly, the discretion is driven by context. And the context is very hard to program. How do you say "you take the trade if its early afternoon, has tested a significant S&R cleanly and ...." to your computer. What most seem to do is build systems that don't do that and as a result you have higher drawdowns ... but in return you don't have to sit and filter for context before every entry.

I experienced similar problems when i started out also.
Not quite sure what elements you would be analysing early afternoon so cannot comment. As for the support and resistance, it depends on how clearly and what method you use to deifine it. Discretionary methods by drawing two lines on a chart by randomly selecting two points will be difficult to program. The clearer and more logical your identification process the easier it is.

 

[Citizen2007]

Mechanical Trading

Most mechanical systems are right less than 50% of the time, most in the 35% to 45% range, this means you will suffer long losing streaks at times. Maybe 10 losers in row, sometimes you might only get 2 small winners in 20 trades.

Not sure if I agree entirely with the percentages. Obviously these can vary depending on your initial strategy . But is it not easier to prevent the losers mechanically by adding filters, which you could easily overlook manually?

As for taking every trade, this can be filtered out by defining certain conditions with a set value, eg using MACD 60m crossedup + 15m crossed down + 5m Crossed down, this combination can be given a specific value and you can decide whether to take it or trade it.

 

[Adamus]

Not sure if I agree entirely with the percentages. Obviously these can vary depending on your initial strategy . But is it not easier to prevent the losers mechanically by adding filters, which you could easily overlook manually

no, adding filters will gobble up your degrees of freedom and curve fit your system to the past.

 

[NetTecture]

Not necesarily, and only if your filters are of a curfe fittet nature.

For example, one generally accepted filter is using multiple fime frames. This gets pretty graphy pretty fast, and hard to do manually. 2-3 timeframes per instrument mean you can only trade view.

A computer has no prob lem doing that on 20 instruments. Or 200.

And it does not generally fall into "curve fitting".

Same with some general trend direction filter.

The question is whether one cuts down the right degree of freedom or not

 

[Citizen2007]

The filters are not designed to curve fit, but to identify why the same conditions did not succeed this time. Every system has to have a degree of self analysis, it may not quite reach 100% accuracy but many losing trades can be eliminated using multi timeframe/price/indicator analysis in different ways.

Like any manual system, it is also essential to minimise losing trades in an automated system. Thus understanding your system is priority number one.

 

[StratOpt]

To answer the question, I have been able to run with fully automated systems since around 2001 so it is entirely possible to do so. I am certain that I am not the only person doing this. I think the biggest keys to doing so successfully will be with a very good understanding of both the strategy development and the strategy testing environments. Once you understand and employ the proper methods then you will find that your success rate in development of such strategies will go up tremendously

 

[fantastic]

To answer the question, I have been able to run with fully automated systems since around 2001 so it is entirely possible to do so. I am certain that I am not the only person doing this. I think the biggest keys to doing so successfully will be with a very good understanding of both the strategy development and the strategy testing environments. Once you understand and employ the proper methods then you will find that your success rate in development of such strategies will go up tremendously

I am curious to know what is your return on capital / year over these 8 years.
Do you do more than 48%?

 

[StratOpt]

My average annual return on total capital has been 44% over those years so no I don't meet your average of over 48%. I do not consider money held in the account for margin purposes, but rather a full "traded against" account. Return on margin account would be much higher.

 

[Chorlton]

Most mechanical systems are right less than 50% of the time, most in the 35% to 45% range, this means you will suffer long losing streaks at times. Maybe 10 losers in row, sometimes you might only get 2 small winners in 20 trades.

You need to do lots of backtesting and have lots of discipline and you cant second guess the system either. You have to take every trade, you never know which ones will be the big winners.

I am guessing that your stats on mechanical systems are based on medium to long term trend following type strategies?

Trend-Following strategies do seem to offer low win rates (around 40% is not uncommon) but the R:R is always relatively high and this combination is what makes them overall profitable.

In contrast, short-term Mean Reversion systems can offer high win rates but with low R:R but again can still be overall profitable.

However, BOTH approaches can be mechanically coded and traded.

Just my 2 cents.....

Chorlton

 

[james1234]

i have through my own testing found that there seems to be more money in the trend following. I have tested scalp settings and also mid range settings but always my best performance comes from a lower win rate and just one trade a day. I was and still am a scalper by nature but since i started my backtesting i have conclusivesly found trend to be better in the long run and now i have to adapt which is gonna be hard as im a scalper at heart.

 

[fantastic]

My average annual return on total capital has been 44% over those years so no I don't meet your average of over 48%. I do not consider money held in the account for margin purposes, but rather a full "traded against" account. Return on margin account would be much higher.

This is near the top returns as it seems the top audited/documented auto trader has been averaging 48% a year for the last 11 years! Achieved by a Ivy league math professor who takes 20% of the profits on top of his fix salary.

What is your background? were you a successful manual trader who happened to have a top scientific and IT background as this guy?

 

[StratOpt]

I am a computer programmer who was mentored in the trading game by a very smart gentleman who had been operating in the stock markets for the previous 14 years. I studied hard and learned well and was able to take what I had learned and apply it to the automated trading arena. I was originally hired by this gentleman to create computer programs and to do the trading for him. I worked for him until he retired fully. The performance values given were what I achieved with the automated strategies and framework of trading that we developed.

 

[littlemog]

that's fantastic returns to me.

how many years have you managed to hit this %?

I am a computer programmer who was mentored in the trading game by a very smart gentleman who had been operating in the stock markets for the previous 14 years. I studied hard and learned well and was able to take what I had learned and apply it to the automated trading arena. I was originally hired by this gentleman to create computer programs and to do the trading for him. I worked for him until he retired fully. The performance values given were what I achieved with the automated strategies and framework of trading that we developed.

 

[GCC]

I am yearning for a 100 % mechanical trading system that is profitable?. Is this possible?

I think it all comes down to what you mean by "mechanical".

If you want to be trading successfully, you have to "mechanically" follow your rules. I'll give myself as an example, I follow the following procedures mechanical for decision making:

1. Find out trendlines and S/R levels (including fibonacci).
2. Watch the candlestick patterns to see how the price reacts to those lines.
3. If there is a clear signal, I enter a trade.

I follow these rules religiously as in a sense I am trading "mechanically". The point is many of these "mechanical" stages are subjective. For example, how can you define a certain candlestick pattern and also its "reaction" towards the trendlines? And S/R lines usually refer to an abstract level of price rather than a rigid line.

So, although the steps are mechanical, the decision making process is not made up of black and white, yes or no rules, so that the decision process depends on experience and cannot rely on "mechanical" processing formula alone. Trading, after all, is a probabilistic game. While a human may take into account of change in social mood or other subtle factors to modify his "mechanical" method, a computer has a harder time in this self-adjusting, self-learning process. Therefore, as another member pointed out in another thread, this constant adjustment and fine-tuning of an EA is not really "sexy" after all.

Perhaps the only exception to this is neural network type EAs (connectionist mechanism, for those who study cognitive science). You expose the EA in a certain kind of market, let it learn how to trade it, and then when that kind of market happens again, you employ that EA to trade. Even then, you still have to invest a lot of time for the learning process.