I’m still a newbie trying to understand how to use ranking systems.
I did a simple system with EBITDA/EV and from 2005 to 2016. The ranking result is good, with 90/100 percentile around 15% while the S&P is at 5-6%.
However, when I tried it with a simulated portfolio (5$ trade and 0.5% slippage), I was barely able to get over the benchmark. Buying rule was buying 10 stocks using the ranking system and rebalancing each week (I used " True" as a selling rule).
Your turnover is too high and transaction costs will eat into your profits. Also, use variable slippage as opposed to fixed slippage unless you have a very good grip on what your slippage should be for the stocks that you are trading. Also, make sure that you have set the flag to allow stocks to be rebought at current rebalance (if sold).
In addition to Steve and Walter’s suggestion I would add that any ranking should be a combination of factors, not just one factor alone. A single factor can recommend stocks with extreme values which are no guarantee for future outperformance.
A good way to start is to use one of the p123 ranking systems and try to improve from there.
Florian
PS: another important factor is the liquidity of a stock (the higher the better, and the lower will your slippage become).
wwasilev: Thank you very much for the explanation of variable slippage. It seems like a nice idea to use it.
sevensisters: Yes I know there should be many factors, but I want to try some strategy first, to see if it would work. I am not sure about this comment, however: “A single factor can recommend stocks with extreme values which are no guarantee for future outperformance.” I don’t understand how having one or two or three factors can change anything about the likelyhood of past performance to produce future outperformance. Let’s say I use EV/EBITDA, rank the result, and use that, if it worked well from 1999 to 2016, shouldn’t it work well in the future as well?
tjbirnie: Thanks! I used rank <98, rank <97 and rank <96 and it helped!
There must be some kind of catch. Using EV/EBITDA ranking and variable slippage, universe = S&P 500, I get a 15% annual return in simulated portfolio from 1999 to 2016. I ran a rolling test and results were good as well, over the benchmark all the time.
Seems easy until you live through that 71% drawdown (as in your sim) or spend the last 2 years (2014-now) with no net gain wondering if there has been a permanent change in the market.
But does it work? Would it produce the same results in real market conditions?
I agree that 71% drawdown is terrible, but the S&P drawdown was ugly as well, and for the last 1.5 years, S&P has been going nowhere as well (well, excepted since maybe last month).
I am not sure I understand your point. As I said, I am just a newbie. I am only trying to understand what is wrong with this model.
I have another question as well. I use ranking to buy the 10 best according to EBIT/EV. It works very well (companies are bought at rank 99.X and solder under 98) if I add no more rules. But if I add rules like sector!=FINANC // no Financial companies or sector!=UTILIT // no utilities it doesn’t work as well. For some reason stocks are bought because of “Buy/Sell Difference” (what is that?!) and a single stock (TSO) got bought three weeks in a row without selling. Why is that?
Nothing is wrong with the model. You may want to diversify. You may even find better models as you go. Actually, Wes Gray might agree with you. His book Quantitative Value has much the same model. He spends a lot of time in his book on an additional factor that adds little–at least in the backtests that I have done. His book has a complex way of identifying stocks at high risk (red flags): it is possible that I just did not duplicate this adequately in my backtest. Wes Gray argues that ROC which Greenblatt uses does not add much.
Nothing is wrong your model at all! Wes Gray is really good and if you are already doing what he has done in his book then you are off to a good start!!!
bastringue,
I guess TSO is bought to get an equal amount on each position again. You should check the configuration of your simulation, if you want to suppress so small rebalances.
If you use the buy-rules, you should know that the rules are applied after ranking. For appling them before ranking you can edit your universe. Usually this results in less turnover and better performance.
Jrinne: Thanks… How do you think I could diversify? I read Tobias Carlisle’s book and that’s how I came to believe that ROIC doesn’t add that much (same as you say… I don’t know Wes Gray).
Sebastian: I don’t know how to do that. I deleted the sector!=FINANC part and it works well that way, but I’d like to try without some sectors without having this kind of bug. I don’t understand why the model buys like 19 or 34 shares of TSO. I chose up to 50% from the equal positions and it still does that. Don’t know why.
How do you edit the universe to apply the rules before ranking?
Also, is it possible to exclude a sector in the ranking system without having to test it in a simulated portfolio ?
You could make a book of systems that are not correlated to reduce the drawdown. Here, for example, is a system that has about the same return but less drawdown. In a book it may reduce your drawdown and increase your Sharpe Ratio.
Even if one of your models is just bad the others may pick up the slack.
No secret to this one. It is just my interpretation of the Hemmerling’s Walkthrough Guide in the help section. Quality stocks with high yield can do less badly in down markets. The ranking system is Marc’s: publicly available.
Help>Tutorials>Walk Through Guide. On this site. But this is just an example that you probably will not use, ultimately. Something I happened to be looking at recently in a book: it did reduce my drawdowns. I’m not using it now.
Reducing the volatility (Beta) should improve Alpha shouldn’t it?