Fully automated futures trading
- Thread starter globalarbtrader
- Start date
I found a strange phenomena when looking at 10-year rolling window Sharpe and Adjusted Sortino* ratios.
I only plotted windows that start on 1980 January (discarding 1970s due to its out of whack volatility and low number of futures markets back then in general).
e.g. "198912" on x axis just means end (1989 Dec) of 10-year window that started 1980 Jan.
When adjusted Sortino is higher than Sharpe, it indicates positive skew. When they're equal - no skew.
The interesting thing is that it has always been higher than Sharpe, until 2008 November after which positive skew seems to have disappeared (2008-11 to 2018-10 seems to be the first 10 year period where Adj. Sortino ~= Sharpe). And it remains so to this day. I'm curious as to why this happened? Anything special about ~2008 November? Something to do with post-GFC?
* adjusted Sortino is just Sortino divided by square root of 2, to make it comparable to Sharpe.
I only plotted windows that start on 1980 January (discarding 1970s due to its out of whack volatility and low number of futures markets back then in general).
e.g. "198912" on x axis just means end (1989 Dec) of 10-year window that started 1980 Jan.
When adjusted Sortino is higher than Sharpe, it indicates positive skew. When they're equal - no skew.
The interesting thing is that it has always been higher than Sharpe, until 2008 November after which positive skew seems to have disappeared (2008-11 to 2018-10 seems to be the first 10 year period where Adj. Sortino ~= Sharpe). And it remains so to this day. I'm curious as to why this happened? Anything special about ~2008 November? Something to do with post-GFC?
* adjusted Sortino is just Sortino divided by square root of 2, to make it comparable to Sharpe.
Check this out
It's the average absolute forecast, averaged across instruments. I don't personally think the 1970's are especially trendy. Well the clothes weren't, anyway.
GAT
I just plotted the rolling (daily) skew directly
The one year is in blue, ten years in orange
It does look like there is a gradual downward trend, but I'm not seeing the same drop in 2018. Something bad happens in 2018, when there's the horrible drawdown (and black friday shows a dip as well, but not as bad).
I also plotted the rolling *monthly* skew. That also shows (on the 10 year view) a gradual downward trend until December 1990, and then drops sharply, and then goes down gradually again.
My theory on the downward trend is that it's due to the gradual introduction of more equity and other financial markets which tend to have worse skew, but this is quite a deep problem requiring further research.
GAT
I forgot to specify that I used monthly data to calculate Sharpe and Adjusted Sortino ratios. I have generated the monthly data by taking daily data and compounding it.
The skew disappears around year 1990 in your set. I don't know how to explain the difference now. Perhaps the difference of sortino vs sharpe is not exactly skew.
Sortino excludes (zeros) positive returns when calculating volatility, basically return = min(return, 0). Divisor is still n. If we assume that both positive and negative returns have equal volatility (and 50/50 chance) - no skew - then multiplying the average by 2, or equivalently the stdev by sqrt(2), we should get approximately the normal volatility that is used in Sharpe ratio calculation. Therefore Sortino divided by sqrt(2) should be higher than Sharpe ratio if upside volatility is higher than downside volatility, and approximately equal if upside and downside volatility is similar ("no skew"). Would you agree?
Although there is another possibly important detail in my formulas. For Sharpe ratio I use mean return, for Sortino I use CAGR. However I look at another ratio - CAGR/Volatility, and the pattern is similar there too (meaning CAGR/Vola becomes almost equal to adjusted Sortino since 2008 November).