QuantConnect Community

Nice algorithm!

The code seems right; I didn't come up with some detail that worth to make a change.

The 73% win rate is great, the problem is the drawdown.

THanks JayJayD.  I modified it to improve returns although it didn't help with the drawdown, check this one out.  Any ideas how I can simulate this since the ETF hasn't been around before 2010 and I'd like to see how it works with a bear market?  Any suggestions on reducing the drawdown?  12x return in about 7 years.  The seasonal trading STS method should reduce the drawdowns a lot compared to just hoding the UDOW.  Also if you simulate the first project all the way back to 2000 it does pretty well through both downturns.

> Any ideas how I can simulate this since the ETF hasn't been around before 2010 and I'd like to see how it works with a bear market?

What do you mean by that? isn't the data available as security? If that's the case you can try with Qandl.

> Any suggestions on reducing the drawdown?

Use some kind of risk manager. I implemented one for Forex, but IIRC it can be used with stocks, the code is available in the DailyFx algorithm shared in broadcast.

UDOW has 3x leverage implied, but if trading on IB they will count this against your margin as if you had been 3x DOW yourself (and max portfolio leverage they permit overnight is 2x...).

Keep in mind these leveraged ETFs rebalance daily and so will suffer from decay where, even if the underlying instrument is flat after an up move cancelling a down move, the ETF will be down. If you intend to hold long term, it might be better to hold a non-leveraged ETF that you can rebalance manually using your portfolio margin.

Also, I wouldn't recommend tweaking algo for higher returns on the same data unless you have a sound explanation for why a given tweak should perform better. That is the path to curve fitting/over-optimization, which is arguably the most common problem in trading system development.

As for risk management, it is also a bit of a philosophical discussion as well as having technical aspects.

• There is an optimal leverage given market volatility and draw down characteristics that maximizes return. Assuming you have a representative sample series, you can find the optimal leverage factor for it.
• If you're using a strategy that simply aims to capture market beta (which is pretty close to your case), simply being fully invested when the alpha signal indicates it is likely sufficient, because it also means all your capital is put to work, and assuming you're not leveraged, the worst case scenario is limited anyhow. When you're doing a large number of trades in a volatile asset, all while being leveraged, the question also becomes "what is the maximum amount of money you're willing to lose in a single trade?", for which said risk management code is necessary. Beware though, stoplosses can fail during Black Swan events...

Thanks JayJayD and Petter for your comments.  This method was written in 1999 by Sy Harding called "Seasonal Timing Strategy" (STS).  He wrote the book "Riding the Bear" that was published in 1999.  What I find amazing is that since then the returns with such a simple method are pretty darned good (I'm using it for my IRA).  See backtest below.  This is just with DIA and unfortunately the ETF i'm using for 'bond' in the off season wasn't around till 2003 so it would do better than this i just didn't have an easy fund to put it in back then.  Also the 'drawdown' I don't think is a totally fair comparison of risk.  If you look at each years performance, the STS method never lost more than about 7% in any year back to 2000.  It is true you are fully exposed to the full volatility of DIA for 6 months, but those are the favorable 6 months, and you get much higher returns being OUT of the market the unfavorable 6 months and drawdowns in favorable seasons seem to rebound quickly.  Know of any easy methods that would work better or similar to this, I'd love to pair this with another method that combined might reduce drawdowns and risk further.

Also if this algorithm is only better than 46% of the community and this is ranked 238,137 , and my other algorithms that do very poorly are ranked more like 44%, so i'm not sure how the ranking is working.  And creating an algorithm to backtest knowing we were already in a long bull run seems to me can't be fairly compared to an algorithm like STS which has been around before 2000.  How does one truly rate their algorithm fairly?

amazing @Robert Hillman

The ranking formula is fairly arbitary; but combines sharpe and drawdown. The rank is "238,137th equal" -- there are millions of algorithms.

Its also fairly sharply skewed so a small improvement should help your rank jump. Really though you should judge your algorithm based on your own financial goals.