So, in the spirit of 'keep it simple', I have a strategy that uses 2 ETFs, trades long-only, and can be traded and backtested using publicly-available indices and data.
I hope you don't mind another follow-up question? I ran the code in Spyder and dumped the data to check a few trades. Currently it looks like the trades exit quite frequently when the >15 and <10 VIX1D qualifiers are no longer valid. I wondered how your strategy performs if:
1. You only exit the SVIX long when the band declines below 10 and switch immediately into long VIXY. Vice versa above 15.
2. The VIX3m-VIX qualifier remains in place as per the current strategy.
It looks OK, although I haven't been able to quantify and just eyeballing it. Did you look at this variant yourself? Cheers!
I don't mind follow-up questions, keep 'em coming.
I just looked at a backtest of the variants you proposed. The 10- or 15-threshold with a switch between SVIX or VIXY is similarly profitable, but it doesn't add to annual returns and the returns have a much lower Sharpe from increased volatility.
Also, the strategy you propose would be invested for about 93% of trading days during the period, instead of 47% of trading days. With the strategy as is, you could probably find another strategy that's profitable with that part of your portfolio when VIX1D is between 10 and 15.
Nice analysis, thx for sharing. May be using a more large sample could be a good idea by simulating a VIX1D over the day where 0DTE partially existed (monday, wednesday and fridray, or may be another config). Furthermore, did you tried to calibrate the size of the allocation by regarding some vol ratio over the last week? Or something else?
I was able to get some historical data from Financial Modeling Prep on VIX1D values back until about 2022-04. But I kept them out of this study because the VIX1D wasn't officially available as an index by CBOE, so it kinda felt like cheating. I think the 2022-04 to 2023-04 return was something like 60%-70%, mostly from SVIX's returns. Volatility targeting and simulating on MWF data is also something I could also add in a later update.
I'll admit I somewhat arbitrarily decided to put the strategy at a consistent 10% in the portfolio going forward, since more felt too big for a single position. If I'm calculating it correctly, Optimal Kelly Criterion would use ~40% of the portfolio for this strategy, so I'm at ~25% Kelly now.
This is an amazing intro write up thanks for sharing it must’ve taken some time to put all of this together
Great chart Good work.
Very clean and insightful work. I loved reading the embedded Shannon's Demon paper too.
Can I check what you see for drawdown of the strategy? Looks like a July 2024 long SVIX position will have caused the most damage?
thank you WS!
I just checked into the drawdowns for the strategy, (using yfinance's data to be quick)
Depending on whether you use SVIX or SVXY for the short-Vol section:
- The SVIX-using ideal strategy has a ~28.5% max drawdown,
- The SVXY-using ideal strategy has a ~15% max drawdown,
For comparison, SPY has an ~18% drawdown during the period studied.
Hi LQ,
Thanks vm for the reply.
I hope you don't mind another follow-up question? I ran the code in Spyder and dumped the data to check a few trades. Currently it looks like the trades exit quite frequently when the >15 and <10 VIX1D qualifiers are no longer valid. I wondered how your strategy performs if:
1. You only exit the SVIX long when the band declines below 10 and switch immediately into long VIXY. Vice versa above 15.
2. The VIX3m-VIX qualifier remains in place as per the current strategy.
It looks OK, although I haven't been able to quantify and just eyeballing it. Did you look at this variant yourself? Cheers!
Hi WS,
I don't mind follow-up questions, keep 'em coming.
I just looked at a backtest of the variants you proposed. The 10- or 15-threshold with a switch between SVIX or VIXY is similarly profitable, but it doesn't add to annual returns and the returns have a much lower Sharpe from increased volatility.
Also, the strategy you propose would be invested for about 93% of trading days during the period, instead of 47% of trading days. With the strategy as is, you could probably find another strategy that's profitable with that part of your portfolio when VIX1D is between 10 and 15.
Hi LQ,
Thanks for this, very insightful, cheers.
Nice analysis, thx for sharing. May be using a more large sample could be a good idea by simulating a VIX1D over the day where 0DTE partially existed (monday, wednesday and fridray, or may be another config). Furthermore, did you tried to calibrate the size of the allocation by regarding some vol ratio over the last week? Or something else?
thx a lot Lay, nice to see you again
Hi quantawolfy, I'm glad to be back as well.
I was able to get some historical data from Financial Modeling Prep on VIX1D values back until about 2022-04. But I kept them out of this study because the VIX1D wasn't officially available as an index by CBOE, so it kinda felt like cheating. I think the 2022-04 to 2023-04 return was something like 60%-70%, mostly from SVIX's returns. Volatility targeting and simulating on MWF data is also something I could also add in a later update.
I'll admit I somewhat arbitrarily decided to put the strategy at a consistent 10% in the portfolio going forward, since more felt too big for a single position. If I'm calculating it correctly, Optimal Kelly Criterion would use ~40% of the portfolio for this strategy, so I'm at ~25% Kelly now.