my mechanical system
- Thread starter robottrader
- Start date
Looking up the posts of acrary is good advice.
Sounds like you have already complemented one of your system's biggest potential weaknesses - single stock disasters.
Your system looks like a dip-buying method to me. I could be wrong but it doesn't really matter. If you expose 20% of your equity to one stock, there is a small but real chance that it gaps down 50% overnight. There goes 10% of your portfolio.
If your backtest is not considering de-listed stocks then this increases the chance that historically you would have purchased some stock that is no-longer listed and therefore not in your back-test. Why was it delisted? It might be that it went bancrupt (big loses). This you have considered I'm sure.
Lastly, if it is a dip-buyer or uses limits to enter positions - are you using EOD data? If so, you need to check how many signals are generated per day. Why? Because your backtesting platform is likely not choosing which stock you're purchasing realistically. It will likely allocate spare captial on a random basis. e.g. 40% of portfolio is in cash. 10 signals that day. 2 new positions needed. Each stock has a 1/5 chance for selection. In the real world, all limits go in at the start of the day. The weakest stocks will hit limits first and the other limits are pulled. The weakest stocks tend to perform less well after purchse than the slightly stronger stocks that get hit later in the day. The only way to check the effect of this bias is to reduce position size to a point where every trade is being taken (or use intraday data). Please disregard this whole paragraph if you are using intraday data already.
Sounds like you have already complemented one of your system's biggest potential weaknesses - single stock disasters.
Your system looks like a dip-buying method to me. I could be wrong but it doesn't really matter. If you expose 20% of your equity to one stock, there is a small but real chance that it gaps down 50% overnight. There goes 10% of your portfolio.
If your backtest is not considering de-listed stocks then this increases the chance that historically you would have purchased some stock that is no-longer listed and therefore not in your back-test. Why was it delisted? It might be that it went bancrupt (big loses). This you have considered I'm sure.
Lastly, if it is a dip-buyer or uses limits to enter positions - are you using EOD data? If so, you need to check how many signals are generated per day. Why? Because your backtesting platform is likely not choosing which stock you're purchasing realistically. It will likely allocate spare captial on a random basis. e.g. 40% of portfolio is in cash. 10 signals that day. 2 new positions needed. Each stock has a 1/5 chance for selection. In the real world, all limits go in at the start of the day. The weakest stocks will hit limits first and the other limits are pulled. The weakest stocks tend to perform less well after purchse than the slightly stronger stocks that get hit later in the day. The only way to check the effect of this bias is to reduce position size to a point where every trade is being taken (or use intraday data). Please disregard this whole paragraph if you are using intraday data already.
i agree, the stats make this look like a dip buying system.
* I agree with previous poster what is your tie breaker in case of multiple signals
* Are you including in your analysis stocks that trade 100 shares a day (always make systems look great)
* If your market entry exit slippage will be greater than you think (ie if you are using opening/closing prints) especially thin stocks. (ie when ur stop is hit you don't get filled at the exact price or if your market exiting at some point)
* what will YOUR size do to effect the market, that is if your trading thin stocks you might be too big for them.
* survivorship bias this is a big one you aren't including stocks that might make your system fail in the past since they aren't in your database even tho in real life you may have blown up. At the very least it will probably make your draw downs considerably worse.
Your edge is small so the slippage could make this system have a negative expectancy.
* I agree with previous poster what is your tie breaker in case of multiple signals
* Are you including in your analysis stocks that trade 100 shares a day (always make systems look great)
* If your market entry exit slippage will be greater than you think (ie if you are using opening/closing prints) especially thin stocks. (ie when ur stop is hit you don't get filled at the exact price or if your market exiting at some point)
* what will YOUR size do to effect the market, that is if your trading thin stocks you might be too big for them.
* survivorship bias this is a big one you aren't including stocks that might make your system fail in the past since they aren't in your database even tho in real life you may have blown up. At the very least it will probably make your draw downs considerably worse.
Your edge is small so the slippage could make this system have a negative expectancy.
I agree with most of the previous posters, however I only trade mechanical systems so i have faith in their assumptions.
To answer your question - float = initial capital. A fixed float removes all profits when allocation trade size. For this system you would only have 200k trade sizes. Therefore, your 20% is based on a fixed capital amount.
Considering this trade size, are your tickers liquid enough. I would suggest limiting your trade size to 1% of (10 day) average volume OR filtering out tickers with an (10 day) average volume < 20m.
As for multiple alerts for dip buyers, i have found that if you rank by volatility or liquidity (and only taking the highest alerts) is quite useful. eg - If you only have 40% capital available you would place orders for the 2 alerts with the highest volatility.
pneuma
To answer your question - float = initial capital. A fixed float removes all profits when allocation trade size. For this system you would only have 200k trade sizes. Therefore, your 20% is based on a fixed capital amount.
Considering this trade size, are your tickers liquid enough. I would suggest limiting your trade size to 1% of (10 day) average volume OR filtering out tickers with an (10 day) average volume < 20m.
As for multiple alerts for dip buyers, i have found that if you rank by volatility or liquidity (and only taking the highest alerts) is quite useful. eg - If you only have 40% capital available you would place orders for the 2 alerts with the highest volatility.
pneuma
return max DD
2001: 79%, 14%
2002: 65%, 11%
2003: 66%, 7.1%
2004: 73%, 4.7%
2005: 55%, 6.9%
2006: 55%, 5.5%
2007: 104%, 4.5%
Can you share some of your experience trading with real-money? and what criteria you look at of a mechanical system before starting trading?
i filter out stocks trading less than $10M/day.
this is so true!!! i have a filter to drop those with small volatility to reduce the number of candidates.
agree... the real DD in 2001 and 2002 could be a lot worse. On the other hand, acquisition will also makes a stock delisted and in the last 6 months, my system happens to pick 2 with huge profit.
I trade about 7 mechanical systems. I have a "main account" with which trades 4 systems with uncorrelated performance metrics using significant leverage. I find that multiple systems designed for different market conditions work the best when working together.
I have other accounts which i trade a couple of trend following long systems - more of a TA based active investing approach.
Humm, what can i say. I trade mechanical systems because my computer generates the signals nightly and i can do other things during the day. I use automated intraday entries and exits. I can't think of anything worse than day trading.
You should expect to loose more than your expect so, imagine triple your historical max DD; and you must engage the market like a clinical psychopath - without emotion. DO NOT 'CARE' about you trading - focus on the important things like what wine to have with dinner, family etc
It really depends on the system. I have TFL systems with a win:loss of 35%, but they make heaps of cash. I have Dip-Buy systems with a win:loss of 80% that make heaps of cash.
I look something that's really fkn simple (idiot proof) with a good to very good Sharpe ratio (>1.25); enough profit to make it worthwhile; a straight closed equity curve; an Average Annual Return:Max DD >5; and expectancy >3.
Or something like that. Yep i would trade you system after looking at the code.
agree about running multiple different systems.
What's TFL?
Do you design all 7 systems by yourself or got from somewhere? really amazing! I cannot even get 1 working
like your idea of return/maxDD.
Thanks a lot!
What's TFL?
Do you design all 7 systems by yourself or got from somewhere? really amazing! I cannot even get 1 working
like your idea of return/maxDD.
Thanks a lot!