Power Dip System Backtested Performance Results from 1/1/1990 to 11/23/09:
Performance Highlights:
- 71% of trades win with 5396 trades recorded.
- Average trade of 1.15%.
- Maximum System Drawdown of -15.84%.
- Sharpe Ratio of 1.52 using Risk-Free rate of 3%.
- Average trade lasts less than 5 days.
- Exposed to market only 37% of the time.
Parameters Used for Backtests:
- Commissions of .008/share included.
- 10% stop used on all trades.
- 1% of account value risked on every trade.
- 10 maximum positions held at one time.
- All trades opened on the open and closed at the close.
- 10 day average volume requirement of 300,000 shares on only Major Exchange Listed stocks (no OTCBB).
- Minimum daily liquidity requirement of $1,000,000.
- Database from Premium Data.
- Database includes all de-listed issues, minimizing survivorship bias.
Complete Backtest Summary:
I’ll post the equity curve soon, and then we’ll take a look at adjusting the system to suit a trader wanting more risk and higher returns.
I’m hearing that the launch is scheduled for December 3rd.
100k to 38 million in 19 years?
Nice.
Wood,
The Power Dip looks like quite the $$$ making machine there. I do have one question about the backtest run(s):
As the backtest portfolio begins to accumulate equity, does it begin to filter out entries where the position taken is a significant portion of the stock’s “usual” daily volume (if not all or more than 100% of the average volume or liquidity)? Taking a position that is a good percentage of a stock’s share or dollar volume (think low $, low volume stocks that still meet the system’s volume/$ liquidity requirements) could result in serious negative slippage (or lack of fill) that could seriously impact equity growth, particularly on a compounded basis.
KTB
KTB, excellent point. I have not altered the backtest to filter out entries where the position would be a significant portion of the stock’s usual daily volume.
The primary reason I chose not to do this is because I designed this system for me, and other retail traders. I do not have a 7 figure trading account, and I do not expect that most other subscribers to the system would be trading with 7 figures. Therefore, for the average person, there will seldom be a time when they will be significantly impacting performance.
All that being said, each pick is published with the 10 day average volume as well as a liquidity measurement. The picks can be ranked by volume and liquidity so that a trader with a larger account can choose not to take a particular trade if he feel it is not liquid enough.
Also, I have a feeling that many people will not trade the system exactly as it is designed, and will instead choose a method that works for them.
Good point though KTB. I always figured that when my account is so big that it will move stocks, that I will be retired and it will not be an issue 😉
KTB, if I have time over Thanksgiving, I will run the test to exclude trades where the entry is more than x% of a stocks average volume.
Wood,
One suggestion is to use both price and volume for the filter. I have a couple of very similar (short term reversion) systems where I use this. My current filter is dynamic, in that it takes Price * Volume of a stock and compares it to Price * Volume of the SP500. Just a thought…
Hope this helps.
Good Trading…
Hello Bill,
I do use something similar to that. I use (MA(10)*AvgVol(10))>1,000,000 in addition to a straight volume requirement.
What comparison are you making to the S&P? Just a basic ratio?
Thanks Bill!
The average bars held is 3.3 for winners, 6.7 for losers? Perhaps you should close all trade after 5 days 🙂
At one point, I tested an n-day exit. I believe that having all trades closed by the 8-9 day improved things slightly. However, the biggest problem with an n-day exit is that they are extremely curve fit. That is primarily why I do not use them with a long-only system. I may use them with a short-only system, but that is because of the substantial risk one takes on when holding shorts for an extended period of time.
Nice performance stats. Curious, I see a rather large win versus average win. What does the system look like after the top 5% of trades are omitted (based on gain %)? I typically perform this in backtesting to ensure system is not picking up just a few outstanding outliers and skewing the results. Plus, it is very easy to skip a trade now and then due to money, vacations, and/or data issues. So, cutting off the top 5% is a fairly reasonable filter. It is also helpful when tweaking your system…allowing you to mine the majority rather than the minority of winners.
Enjoy your blog. Happy Thanksgiving!
Mike Taylor
Mike,
Given that the backtest is effectively compounding equity over time, the “average” and “largest” values on an absolute $ basis, when compared with each other, this is not going to be an apples to apples comparison. The largest win probably comes from late in the backtest period when equity is near its largest, while the Avg Win is computed by averaging all of the wins from the first $10k positions at Trade 1 through the last, multi-million $ ones at Trade 5396. Comparison of avg and largest win percentages would be more instructive when looking for outliers.
KTB
Mike, KTB is correct about the difference between the large win vs. average win. The difference is great due to the compounding.
We are hitting the road for Thanksgiving travel, so I don’t have much time to address some of these issues right now, but I could make available a spreadsheet containing all of the trades, in the future.
Chopping of the top 5% is not a bad practice, but I’m not sure if I can get AmiBroker to do that. I might need to export the trades and do it manually, and then I would exclude the top 5% of the largest percentage gains.
What I would rather do is run the system without compounding gains, and take a look at that those results. I have done that before, but it was probably a year ago.
Thanks for reading, and Happy Thanksgiving to you too!
KTB, good point, the largest gain percentage and date traded would be useful in the sheet above. That way, you wouldn’t have to assume anything. I do agree that it most likely occurred very early in the simulation. But, it is not always due to the success of the system or timing…can also be data issues. Especially, starting back in the early 90’s. Not a knock on your data…I’ve used them all…and all have their issues.
That’s why it is always important to cap the top x% of trades when evaluating a system. So, you can evaluate the heart of the trades and avoid looking at a perfect world situation.
Woodshedder, I’m not familiar with Amibroker but you’re right, most products won’t handle this type of function…definitely a roll your own situation.
MT
Nice blog Mike.
What happens when you include slippage or enter into trade 1 day later?
b
Blink, why would I enter a trade a day late?
I’m not sure why that would be a valuable consideration, but I ran it anyway. Waiting a day later to enter reduces the CAGR to 27.14%
Introducing slippage of .02 on both sides of the trade lowers the CAGR to 21.01%
I usually test my systems also with 1 day entry/exit delay, to see if it’s not curve fitted. And of course there’s slippage which can greatly reduce returns.
Slippage sucks. I’m working on a project that should allow all of us backtesters to get a better idea of how much slippage can affect trading the open. More info soon on that.
Couple of comments questions for you Wood.
Do you have a comparison of your results with delisted vs without delisted. Just curious how valuable your data was for this particular strategy.
How long did it take to run the backtest on all of this data?
Lastly, nice work as always.
Red, I was pretty stupid about implementing the delisted data. I should have ran backtests across all my systems one last time, and then loaded the delisted data, and ran them again, recording the differences. Unfortunately, I didn’t. All I can tell you is that every system I have ran over the new database which includes delisted data, has improved. This is due primarily to one reason: more opportunity. I will see if I can exclude the de-listed data, in order to give you more specific results.
A friend of mine also implemented delisted data recently, and he ran the Power Dip over it. He trades it at 2% risk, and I believe that testing over the delisted data improved the system around 3% over the test period.
Short systems seem to improve more-so than long-only systems, for what should be an obvious reason.
It takes less than 2 minutes to run all this data in AmiBroker.