A summary is below the sheet.
Click on spreadsheet to enlarge…
November will mark the one year anniversary of the inception of PDS subscription service. As we approach the beginning of the second year, I am in the process of reviewing the system in order to make any necessary changes for implementation the 2nd year. I’m not anticipating any changes to the original algorithm; rather, the changes will likely involve an adaptive position-sizing method, or at the least, refining the original methods.
The tests represented above attempt to answer two questions:
1. Would adding an additional uptrend requirement and trading only low or high volatility stocks improve the system?
2. Does ATR position-sizing or fixed-stop position-sizing work better during bear markets?
Reading the Sheet:
All tests include .01/share for commissions and were run over de-listed data from Premium Data.
- The top half of the sheet (above the blue horizontal line) shows the results of testing from 7/1/2007 – 9/4/2010. I consider this period to encompass the current bear market (which includes the bull run of 2009).
- The bottom half of the sheet (below the blue horizontal line) shows the results of testing from 4/1/2000 – 6/1/2003. This period represents the bear market that followed the tech crash.
- The white columns show the fixed-stop results (1% Risk, 10% Stop) with the uptrend requirement (U) and either low (L) or high (H) volatility stocks.
- The gray columns show the same tests with ATR position-sizing using 1% Risk.
Interpreting the Results:
For this type of system, opportunity is everything. I cannot emphasize that enough.
- While the U_H (uptrend and high volatility) models have a significantly higher average trade during the 2000 bear market period than the baseline models, opportunity is limited to the extent that they under perform in terms of net percentage gain. However, the higher average trade is encouraging and it may mean that we flag those higher volatility picks so that subscribers can adjust position-sizing in an attempt to capture a larger gain. The data show other added benefits to the U_H and U_L picks.
- During bear markets, ATR position-sizing appears to under perform fixed-stop position-sizing. This is contrary to my own intuition. While I need to run similar tests during bull markets to be sure, I’m fairly sure that ATR position-sizing is most beneficial when volatility is low (as we would expect during bull markets) as it requires taking larger positions than we would with fixed-stop sizing. Excluding net percentage gain, the data show that there are other benefits to using ATR position-sizing rather than fixed-stop.
More Testing Ahead…
I am currently working on an adaptive model which will change both risked amount and position-sizing models based on the market environment. Also, I have combined ATR position-sizing with fixed-stops, and the results are encouraging. I will publish the results as soon as they are available.
Don’t have time to read it right now,but I will tomorrow.
Keep up the good work and thanks!
Good stuff woodshedder. It’s great that you run and share all these tests. In a way nothing gives you confidence in an idea like a good backtest.
Very interesting results. One thing I would be curious about is the effect of just the uptrend filter alone, without the volatility filter, or the volatility filter alone, without the uptrend filter.
Another thing I would be curious about would be the effect of reducing the risk/trade (keeping all else constant) in bear markets with the ATR model. In general, it seems like increasing the risk/trade increases the returns without a proportional increase in risk. But I am not sure this would hold for volatile bear markets. I noticed that the 2% risk, 3 ATR model can buy in at a market top, filling up the portfolio, only to see things go pearshaped. Then at the dip, you can’t buy in at the lower price. With a lower risk/trade, you wouldn’t fill up the portfolio at the higher level; therefore, you wouldn’t suffer such large losses at the top and you would be able to benefit from buying the dip. So my hypothesis would be that risk-adjusted returns may be better at lower risk/trade levels during bear markets, while the opposite would be true in bull markets.
Data, glad these things are helpful.
When I test using logic such as “Uptrend with High Volatility OR Uptrend with Low Volatility AND the rest of the algorithm”, results are pretty terrible. Theoretically since a stock will be either high or low volatility, this should capture all of the stocks and add in the additional uptrend factor.
Re: Risk/Trades, I believe your theory is correct and will be borne out by the results. However, I also believe that net percentage gain will be smaller since less was risked, although other metrics will improve, like risk-adjusted.
For these tests I did not limit the number of new buys that could be made in one day. If the portfolio had room for 10 stocks and there were 10 picks, it bought all of them. Limiting number of new buys in one day should slightly improve the CAR/MDD.
Hi Wood,
Just had a thought about your limiting the number of picks/day. Instead of limiting the number of picks, could you limit the proportion of total cash deployed in one day? Maybe something that says if you’re fully in cash on day 0, on day 1 you can only use 50% of your cash max, then on day 2, you can invest the rest of your cash.
Interesting that the uptrend filter doesn’t seem to work by itself, but that it would work better with the volatility filter. That’s somewhat counterintuitive, but then, it seems like this is a pretty complex system with a lot of interacting variables so things aren’t always going to be intuitive and predictable.
Data, with the percent-risk models and fixed stop, it does limit the proportion of cash per day. 5 picks per day at 1% risk means that as much as 50% of the account can be put to use in one day. 3 picks per day at 2% risk means as much as 60% of the account can be put to use in one day.
I’m not sure why they additional uptrend filter doesn’t do well. I’m going to have to look more into it. However, the PDS is really not at all complicated. It is very simple. If you do not count the volume and liquidity filters, all you have is a one factor entry signal, and one factor exit signal, and an objective measure of what constitutes an uptrend.
Thanks for thoughts, Wood. Didn’t think of that, but you’re right in that percent-risk model + fixed stop + max number of positions entered per day = limiting daily cash deployment. I wonder, though, if you could also use a % cash deployed/day cap with the ATR models (since those models tend to perform better otherwise). With the ATR models, the simple number of positions entered/day cap may not stop you from going all in on any single day; even when capping the number of positions entered/day at 1, a pick like the Vanguard Bond Fund (low volatility, high share price) could still use up all the cash.
I am not sure this would actually improve performance, though. It could improve performance if it causes you to avoid getting “trapped” unnecessarily in the bad picks when market tops; or, it could hurt performance if there are many PDS “jackpot” days when it really pays to go all in and you can’t. I suppose the understanding what will happen as you manipulate things is what seems quite complex to me, even though the system is pretty simple – it seems I can come up with possible positives and negatives for any manipulation. I suppose that’s the beauty of the backtest — you can just throw it all in the mix and find out what happens 🙂
Wood,
Have you ever created any statistically indexes for your PowerDip System using Amibroker’s Addtocomposite function? For example, an index that counts the number of stocks that would be long the PowerDip System (as a percent of the total number of stocks) or some other derived piece of data. Then maybe using this as an indicator.
In other words, if a large percentage of stocks are triggering your system perhaps it would be bullish or conversely bearish. Just a thought.
Red, I sure do. I just publish it on the subscriber site, a couple times a month. It is a pretty decent indicator for marking bottoms, but doesn’t work as well for tops.
Looking at the results, the “Max trade % drawdown” is a lot more than the 10% stop. Is that due to overnight gaps down? And could you please explain the ulcer index? thanks
Dave, I’ll give you a good explanation of both when I get back home this evening.