4 ms·

The problem is no one can calculate a Sharpe ratio for crypto. The variance is not well understood.

chollida1 4y ago
Sharpe ratio uses a backward looking variance as it tells you how you traded wrt to the volatility. I assure you, we can trivially look back to see the variance. I mean, how could we calculate a sharpe without knowing the return and volatility, we always use historical for both, its one measure of how we track portfolio returns, which again, are backward looking. Though sharpe isn't used as much as it was 15-20 years go due to it penalizing volatility in positive returns as much as it penalizes volatility in losses.
- queuebert 4y ago
  That is a backward/ex-post Sharpe ratio. Bankroll management requires knowing the forward Sharpe ratio for VAR. You can't know VAR without knowing the forward expected variance. This is why black swan events wipe out traders who think they know their risk but really don't.
  - chollida1 4y ago
    Hmm I worked in a bank and ran these calcs and we never use forward variances as you can’t know it for any instrument and you can’t know your return as well. Crypto has nothing to do with this. Are you certain if your facts here because something doesn’t seem right. VAR makes abut more sense but still uses a backward looking variance. Sharpe never uses a forward lookingvariance as this makes no sense as you don’t know your returns ahead of time unless you are Madoff And for VAR we either typically use historical VAR or Monte Carlo, again because you never know your returns ahead of time so trying to do any risk measure with estimates returns is useless
    - queuebert 4y ago
      This is why banks fail. Just kidding.... :-P Estimated Sharpe for a trade would be what you think the return should be (e.g. fair market value - current price) divided by the estimated future variance. This is what you estimate for VAR and compare to your risk tolerance. The eventual accuracy of the estimates will determine whether it's a ho-hum trade or a black swan that wipes you out. Black swans are essentially situations in which variance estimates were completely wrong (as opposed to return estimates). Variance is a function of a bunch of things (and correlated with every damn thing). Simply taking historical variance and assuming it will be the same in the future is the laziest possible solution. Black Swan events have woken people up to platykurtic Gaussians and the fact that many real life distributions aren't even Gaussian. This is why you use Monte Carlo, because it doesn't need to assume a kurtosis or even Gaussianity, but is more computationally intensive, but not terribly so, but also suffers from low sample number at the tails, so it's not that accurate in extreme situations either. An additional red flag is that, if you have to use MC, then you don't know the distribution underlying the process, and if you don't know that there might be other things you don't know. Crypto is one of the newest markets, so we understand a lot less about its extreme conditions and the tails are very uncertain. Even with Monte Carlo I wouldn't trust crypto Sharpes one iota. Sorry if you know all this stuff. Thought I should clarify where we probably actually agree but may be thinking of it differently.

Sharpe ratio uses a backward looking variance as it tells you how you traded wrt to the volatility. I assure you, we can trivially look back to see the variance. I mean, how could we calculate a sharpe without knowing the return and volatility, we always use historical for both, its one measure of how we track portfolio returns, which again, are backward looking. Though sharpe isn't used as much as it was 15-20 years go due to it penalizing volatility in positive returns as much as it penalizes volatility in losses.

queuebert 4y ago
That is a backward/ex-post Sharpe ratio. Bankroll management requires knowing the forward Sharpe ratio for VAR. You can't know VAR without knowing the forward expected variance. This is why black swan events wipe out traders who think they know their risk but really don't.
- chollida1 4y ago
  Hmm I worked in a bank and ran these calcs and we never use forward variances as you can’t know it for any instrument and you can’t know your return as well. Crypto has nothing to do with this. Are you certain if your facts here because something doesn’t seem right. VAR makes abut more sense but still uses a backward looking variance. Sharpe never uses a forward lookingvariance as this makes no sense as you don’t know your returns ahead of time unless you are Madoff And for VAR we either typically use historical VAR or Monte Carlo, again because you never know your returns ahead of time so trying to do any risk measure with estimates returns is useless
  - queuebert 4y ago
    This is why banks fail. Just kidding.... :-P Estimated Sharpe for a trade would be what you think the return should be (e.g. fair market value - current price) divided by the estimated future variance. This is what you estimate for VAR and compare to your risk tolerance. The eventual accuracy of the estimates will determine whether it's a ho-hum trade or a black swan that wipes you out. Black swans are essentially situations in which variance estimates were completely wrong (as opposed to return estimates). Variance is a function of a bunch of things (and correlated with every damn thing). Simply taking historical variance and assuming it will be the same in the future is the laziest possible solution. Black Swan events have woken people up to platykurtic Gaussians and the fact that many real life distributions aren't even Gaussian. This is why you use Monte Carlo, because it doesn't need to assume a kurtosis or even Gaussianity, but is more computationally intensive, but not terribly so, but also suffers from low sample number at the tails, so it's not that accurate in extreme situations either. An additional red flag is that, if you have to use MC, then you don't know the distribution underlying the process, and if you don't know that there might be other things you don't know. Crypto is one of the newest markets, so we understand a lot less about its extreme conditions and the tails are very uncertain. Even with Monte Carlo I wouldn't trust crypto Sharpes one iota. Sorry if you know all this stuff. Thought I should clarify where we probably actually agree but may be thinking of it differently.