My recent research has been on the volatility Exchange Traded Products. My focus has been on long trades using VXX and XIV. Although VXX has a very strong downtrend, I am not a fan of developing short strategies on it due to the huge upside risk. I wrote about XIV here and expressed some of the dangers of trading these ETFs.
XIV has an inception date of 11/30/2010 and VXX inception date is 1/30/2009. Market has been in a strong uptrend during most of this period. Because of this, most people use synthetic data for before their inception to test the bear market scenarios. I will be using the data from Six Figure Investing which has synthetic data back to 2005 for both. A comparison of the daily return of the synthetic vs real data from 2011 to 2014 has a correlation of 0.999+. In the next post, I will investigate if this is good enough.
One big issue with the synthetic data is that we only get a closing price. I am not a fan of trying to enter at the close on the signal day. I prefer to enter at the open or close of the day after the signal day. We will see how each of these different entry times change the results.
Given we only have data back to 2005, this really does not give us enough to break up into in-sample and out-of-sample testing. For me, this is not a big deal because IS vs OSS testing is not a critical test for me. I like to be able to do it but it is not necessary
There are lots of sites with strategies for trading XIV and VXX. My favorite site is Volatility Made Simple. Unfortunately, they have not done a new blog post since January and I hope this does not mean he is no longer updating the site. They have tested lots of strategies. I started with “Brute Force Optimization of VRP Strategy” and modified to what I will show next. I was also curious to see how the strategy has done this year.
This strategy is in either XIV or VXX depending on the difference between the current volatility and VIX. Calculate the MaDiff which is 5-day average of the [VIX index – (10-day Historical Volatility of SPY)].
If MaDiff is greater than zero then go long XIV. If MaDiff is less than zero then go long VXX. Nice and simple.
My Change to the Rules
I wanted to optimize some of the parameters, in particular the ‘zero’ value. But I did not want to simply change that to value because the difference between when volatility is low vs high would not be the same. So I change it to a ratio.
- MAVolRatio = (1 to 8)-day average of (VIX/((2 to 15)-Day HV of SPY))
- If MAVolRatio greater than (.75 to 1.25 in steps of .05) then go long XIV
- If MAVolRatio less than (same value as previous rule), then go long VXX
The spreadsheet will have the full set of the results.
How has it done?
My first question how has this choosen parameters done since he published this in mid-2014. I used a a cut off value of 1, which is the equivalent of zero in his test. The moving average value is 4 and the volatility length is 8.
Click to see larger image.
Wow, it has collapsed. I guess this failed the real life walk forward test. Could have the values picked been unstable ones? Now, so far this year it is doing better. Is it back?
Comparison of Entry/Exit times
I want compare the different results of entering on signal day vs the next open vs next close. Since we are using the open, we have to test from 2011.
We can already see a substantial change in the CAR using the different entry/exit times. With an average hold of 26 market days, I was surprised to see such a difference in results. As expected the drawdowns are huge given you are in the market all the time. But as the Taming High Return and High Risk post pointed out, we do have one method of reducing the drawdown but at expense of the CAR which there is not much of.
Testing back to 2005 with different parameters
Let us look at different parameters.
(Click on image for larger version)
These results still show a poor performance for the last few years. I also don’t like how sensitive the results are to the cut off value. It looks like the search must continue.
Fill the form below to get the spreadsheet which contains all the variations tested and additional statistics.
It was interesting to see a strategy that was doing so well, with relatively simple rules, just collapse over the last few years. Would I trade this? I would not due to the poor performance recently. I would want a higher return given the risk. Even if I used cash to help reduce the drawdowns. I wish the parameters where a little bit more stable. The biggest question I still have is using the synthetic data valid? That will be explored in the next post.
Good Quant Trading,
Fill in for free spreadsheet: