I've been listening to the algorithmic advantage podcast, which is a series run by two (now down to one) Australian traders that interview various algorithmic traders regarding their strategies. The series of guests and the blog posts have been really informative in terms of motivations and strategies.
One of the episodes spun me off onto the turtle traders, which is a fascinating story where a legendary commodity trader made a bet with his partner that he could take a set of people and make them into successful traders using fairly simple rules that he provided. He ended up winning the bet, even though his partner probably was more useful in teaching them what they needed to know.
In essence, the turtle traders were looking for outliers to ride up and down in the commodities market (expanded to currencies). Such opportunities are fairly rare, but it only takes a few hits to make the big profit. This is an example of a positive skew trading strategy, where most of the trades are small losers but the expected return is positive. As I understand it, such strategies work because the markets have fat tails. Because the opportunities are rare, careful position sizing and risk control are vital so that the many losers don't kill the returns.
The commodities market is different from equities and bonds, because it is basically haggling between producers and buyers to come to a mutually advantageous hedging of future risk for relatively inelastic quantities. Equities tend to be tied to underlying fundamentals to a much larger extent, so that commodities tend to be more consistently driven by emotions than equities (in my understanding). IMO this makes commodities/currency trading behave with a larger price component related to shocks driven by emotion.
I think that volatility, cryptocurrencies, and nonfungible tokens could also be fairly characterized as more driven by emotion than equities.
I've been intrigued by the ways that volatility can spike just during market crises, which makes long volatility an attractive hedge. Unfortunately, there is no way to directly invest in volatility. The best that one can do is to use futures, either directly or as ETFs. Long volatility has a tremendous loss rate, so much so that buying and holding cannot really be justified.
The way that long vol spikes during crises (a rapid climb and almost as rapid decay) suggests that a trading approach like the turtle traders used might be interesting. After playing around a bit, it seems that using VIX as a canary for VIXY or UVXY or UVIX (1, 1.5, or 2x long VIX) may be fruitful. The key is to only enter the trade when the end of day VIX is the maximum over some long back duration, like 10 months, then get out when end of day VIX is less than some short duration. The precise entry signal is not very important, but there is some uncertainty on the exit. I hedge in backtests by splitting the initial bet into three exits, representing the 1, 2, and 3 day minimum. The different exits allow rebounding.
The backtest for this is
![Image]()
The lines above one are three portfolios, all consisting of SGOV except when the long vol is active at 1/3, 2/3, or 3/3 of the portfolio (depending on what fraction has exited). The portfolio with VIXY is orange, UVXY is blue, and UVIX is green. The three lines below 1 are VIXY (yellow), UVXY (red), and UVIX (cyan).
The first thing to notice is that buy and hold is a huge loser.
The second thing to notice is that the overall strategy rarely trades, and it's only allocated to long vol on average 1.7% over the entire period (effectively 17 days every four years on average, although a bit more days because it tails out on exit).
These strategies produced CAGR of 12, 14, and 20% (accounting for bid/ask slippage, but probably optimistically). Some of that is SGOV. UVXY tested the best.
There is likely some additional benefit in using this as an overlay, running the normal strategy and replacing the equities with UVXY when the signal hits. Equities will tend to be crashing when a big VIX spike hits (e.g., 2020); replacing a 3x LEFT equity with UVXY at this time will tend to have a dual benefit of replacing some equity drop in favor of the UVXY gain.
Notice that over the 20-year period, there were only 8 or 9 positive events, perhaps 4 neutral events, and 4 or 5 losers. I tried more frequent triggers, 2005 to 2012 it doubled returns but more than gave them back afterwards. Maybe a good time to activate is during grinding downturns and potentially when shocks are signaled (like Covid).
This has got to be an automated thing to actually trade, it happens so rarely that it would be hard to handle manually.
Crypto is also something that seems to be reasonably traded using a similar strategy, especially earlier in the coin's life when it is more emotional. I used just the same approach with GBTC, using the BTC-USD ratio as the canary. It seemed to work pretty well with two entries (3 and 7 days lookback) and two exits (1 and 2 days lookback). This would have had ~90% CAGR from inception with max drawdown of 39% (again, probably optimistic on slippage).
![Image]()
The behavior seems to be a series of jumps followed by decaying decline. Even in the relative calmer last 3 years, CAGR was roughly 23%.
Again, this has to be done with a bot.
One of the episodes spun me off onto the turtle traders, which is a fascinating story where a legendary commodity trader made a bet with his partner that he could take a set of people and make them into successful traders using fairly simple rules that he provided. He ended up winning the bet, even though his partner probably was more useful in teaching them what they needed to know.
In essence, the turtle traders were looking for outliers to ride up and down in the commodities market (expanded to currencies). Such opportunities are fairly rare, but it only takes a few hits to make the big profit. This is an example of a positive skew trading strategy, where most of the trades are small losers but the expected return is positive. As I understand it, such strategies work because the markets have fat tails. Because the opportunities are rare, careful position sizing and risk control are vital so that the many losers don't kill the returns.
The commodities market is different from equities and bonds, because it is basically haggling between producers and buyers to come to a mutually advantageous hedging of future risk for relatively inelastic quantities. Equities tend to be tied to underlying fundamentals to a much larger extent, so that commodities tend to be more consistently driven by emotions than equities (in my understanding). IMO this makes commodities/currency trading behave with a larger price component related to shocks driven by emotion.
I think that volatility, cryptocurrencies, and nonfungible tokens could also be fairly characterized as more driven by emotion than equities.
I've been intrigued by the ways that volatility can spike just during market crises, which makes long volatility an attractive hedge. Unfortunately, there is no way to directly invest in volatility. The best that one can do is to use futures, either directly or as ETFs. Long volatility has a tremendous loss rate, so much so that buying and holding cannot really be justified.
The way that long vol spikes during crises (a rapid climb and almost as rapid decay) suggests that a trading approach like the turtle traders used might be interesting. After playing around a bit, it seems that using VIX as a canary for VIXY or UVXY or UVIX (1, 1.5, or 2x long VIX) may be fruitful. The key is to only enter the trade when the end of day VIX is the maximum over some long back duration, like 10 months, then get out when end of day VIX is less than some short duration. The precise entry signal is not very important, but there is some uncertainty on the exit. I hedge in backtests by splitting the initial bet into three exits, representing the 1, 2, and 3 day minimum. The different exits allow rebounding.
The backtest for this is
The lines above one are three portfolios, all consisting of SGOV except when the long vol is active at 1/3, 2/3, or 3/3 of the portfolio (depending on what fraction has exited). The portfolio with VIXY is orange, UVXY is blue, and UVIX is green. The three lines below 1 are VIXY (yellow), UVXY (red), and UVIX (cyan).
The first thing to notice is that buy and hold is a huge loser.
The second thing to notice is that the overall strategy rarely trades, and it's only allocated to long vol on average 1.7% over the entire period (effectively 17 days every four years on average, although a bit more days because it tails out on exit).
These strategies produced CAGR of 12, 14, and 20% (accounting for bid/ask slippage, but probably optimistically). Some of that is SGOV. UVXY tested the best.
There is likely some additional benefit in using this as an overlay, running the normal strategy and replacing the equities with UVXY when the signal hits. Equities will tend to be crashing when a big VIX spike hits (e.g., 2020); replacing a 3x LEFT equity with UVXY at this time will tend to have a dual benefit of replacing some equity drop in favor of the UVXY gain.
Notice that over the 20-year period, there were only 8 or 9 positive events, perhaps 4 neutral events, and 4 or 5 losers. I tried more frequent triggers, 2005 to 2012 it doubled returns but more than gave them back afterwards. Maybe a good time to activate is during grinding downturns and potentially when shocks are signaled (like Covid).
This has got to be an automated thing to actually trade, it happens so rarely that it would be hard to handle manually.
Crypto is also something that seems to be reasonably traded using a similar strategy, especially earlier in the coin's life when it is more emotional. I used just the same approach with GBTC, using the BTC-USD ratio as the canary. It seemed to work pretty well with two entries (3 and 7 days lookback) and two exits (1 and 2 days lookback). This would have had ~90% CAGR from inception with max drawdown of 39% (again, probably optimistic on slippage).
The behavior seems to be a series of jumps followed by decaying decline. Even in the relative calmer last 3 years, CAGR was roughly 23%.
Again, this has to be done with a bot.
Statistics: Posted by Hydromod — Tue Jan 28, 2025 9:21 pm — Replies 103 — Views 24001