I’m hoping some of you with a little experience can help me out here. I’ve created a ranking system that produces good results on its own. However, when I try to create a sim that uses it, the results are terrible. It’s a basic mean reversion ranking system that has a fairly smooth performance graph, with the 95-100 ranks returning over 28%. The rank was tested with a 1 week rebalance.
[size=1][/size]
Rank code is:
Ret1W%Chg
Ret4W%Chg
My sim rules are very simple. Buy stocks with a rank of 95 or more and sell when when they fall below 70. My commission is set at $2.
How could this be? I’m sure i’m doing something stupid, but I can’t imagine what.
One thought is to set your # of buckets on the histogram to something like 100 or 200 so you can see the very tip of your performance distribution.
1500% turnover seems pretty high to me, but your ranking system must be seeing massive movement of stock rankings on a weekly basis (moving rapidly from the 95% plus percentile to below 70th percentile in short time). It looks like ranking by 1wk and 4wk returns might be generating the turnover. (Stock in top 5% of returns in 1wk or 4wk might be highly volatile in following weeks, and those same stocks might be likely to end up below 70th percentile at next measure.) Makes me wonder if transaction cost and slippage might be eating you up given such high turnover.
I can’t tell, but if the sim is running weekly I could see that increasing turnover also given the 1wk return rule likely being volatile.
Just a few thoughts. Hope it’s maybe helpful
(sorry for the edits - I was just shooting from the hip and fixing it after I hit submit)
edit #4: I’m guessing you might be getting quite a few very small stocks, and slippage on those can be very large. (assuming you’re using variable slippage in the sim settings)
Ranking system performance is without transaction costs (and slippage), sim is with both. This, together with the high turnover, makes the big difference.
These pullback / reversion systems are hard to trade profitably, if at all.
Typically when you build a ranking system that incorporates price and volume factors, you tend to get a lot of turnover. If you created the ranking system with slippage set to 0 then you won’t see the effect of the turn over until you build the trading system and turn on slippage. Personally I like to create a price and vol node in the ranking system and then control what weight I want to give to it after the rest of the ranking system is built. My rule of thumb to never give more than 30% weight to a price and vol root node even if degrades the ranking performance. I don’t mind a little TA but I don’t trust it that much.
The combination of the ‘All Fundamentals’ Universe and variable slippage is killing the sim. That universe choice is terrible since it includes many low-price, low volume issues; about 20% of the stocks have an AvgDailyTot(20,0) of less than $10000. The variable slippage model makes those transactions very expensive.
For fun, try setting your sim slippage to zero. Of course, that model is not tradable in real-life, but you’ll quickly have a trillion dollar portfolio.
Start with a tradable universe, keep variable slippage and rethink the ranking system.
[/size]
[/size]