Serving the Quantitative Finance Community

 
User avatar
outrun
Posts: 4573
Joined: January 1st, 1970, 12:00 am

Re: Nonparametric Dependence Measures in Quantitative Trading

January 9th, 2018, 12:32 pm

I wouldn't want to decompose the P&L into a "success ratio * expectation of positive/negative trades".

A success ratio is only informative if the average P&L of a loss isn't much different as the average P&L of a win. .. but it is well known that P&L is typically asymmetrical and so it will have some unspeficied bias: is 60% good? You can't know. What if your losses -although less frequent- are twice as big on average?

Survival is indeed the basis of everything, but in that context the same argument holds: success ratio is less informative than looking at the P&L distribution across timescales because success ration can't distinguish between strategies that have either small or large random drawdown, and which is turns has a high impact on the survival probability.
 
User avatar
staassis
Topic Author
Posts: 27
Joined: April 12th, 2014, 5:10 pm

Re: Nonparametric Dependence Measures in Quantitative Trading

January 9th, 2018, 3:00 pm

success ratio is less informative than looking at the P&L distribution across timescales because success ration can't distinguish between strategies that have either small or large random drawdown, and which is turns has a high impact on the survival probability.
This is wrong. I can give you an example of a strategy with high expected returns, low success ratio and high drawdown (high maximum continuous loss). Expected returns are not really a measure of risk. That is why drawdown is always mentioned as a separate performance measure when a PM comes for an interview at a hedge fund.  
 
User avatar
Traden4Alpha
Posts: 3300
Joined: September 20th, 2002, 8:30 pm

Re: Nonparametric Dependence Measures in Quantitative Trading

January 10th, 2018, 12:57 pm

success ratio is less informative than looking at the P&L distribution across timescales because success ration can't distinguish between strategies that have either small or large random drawdown, and which is turns has a high impact on the survival probability.
This is wrong. I can give you an example of a strategy with high expected returns, low success ratio and high drawdown (high maximum continuous loss). Expected returns are not really a measure of risk. That is why drawdown is always mentioned as a separate performance measure when a PM comes for an interview at a hedge fund.  
Neither is success ratio a proper measure of risk.  If the losses are always tiny, there's no risk.

Success ratio is an extremely poor quality indicator of trading system performance.  There's too many ways a system with a high win rate can be a horrible money loser and a system with a very low win rate can be a money maker.
 
User avatar
staassis
Topic Author
Posts: 27
Joined: April 12th, 2014, 5:10 pm

Re: Nonparametric Dependence Measures in Quantitative Trading

January 10th, 2018, 2:43 pm

Neither is success ratio a proper measure of risk.  If the losses are always tiny, there's no risk.
Success ratio is an extremely poor quality indicator of trading system performance.  There's too many ways a system with a high win rate can be a horrible money loser and a system with a very low win rate can be a money maker.
Traden4Alpha, you probably have not read the whole thread. My response to Outrun on January 9: "You need to have both1) high success ratio (strong signals) and 2) high Sharpe ratio (return magnitudes that work for you)... However, in my opinion, the success rate is more important. It is highly correlated with your quality as a prop trader. For a prop trader the most important task is to survive. See who survives over the decades. What is the success ratio of Warren Buffet or George Soros? 50.5%? No."
 
User avatar
Traden4Alpha
Posts: 3300
Joined: September 20th, 2002, 8:30 pm

Re: Nonparametric Dependence Measures in Quantitative Trading

January 11th, 2018, 12:52 pm

Neither is success ratio a proper measure of risk.  If the losses are always tiny, there's no risk.
Success ratio is an extremely poor quality indicator of trading system performance.  There's too many ways a system with a high win rate can be a horrible money loser and a system with a very low win rate can be a money maker.
Traden4Alpha, you probably have not read the whole thread. My response to Outrun on January 9: "You need to have both1) high success ratio (strong signals) and 2) high Sharpe ratio (return magnitudes that work for you)... However, in my opinion, the success rate is more important. It is highly correlated with your quality as a prop trader. For a prop trader the most important task is to survive. See who survives over the decades. What is the success ratio of Warren Buffet or George Soros? 50.5%? No."
I read the whole thread.  I've got no idea about the per-trade success ratio of Buffet or Soros.  I do doubt they are representaitve of the entire set of successful traders.

What's interesting is that every trading system has at least two types of signals: the entry signal and the exit signal.  As such a trading system might be good at cutting losers short and letting profits run which creates a large number of tiny losses and a small number of very high returns.  The net result is winner even if most trades are losers.

Survival is important, but that's a function of the magnitude of losses, not the probability of losses.  Again, magntitudes always matter more than probabilities.
 
User avatar
outrun
Posts: 4573
Joined: January 1st, 1970, 12:00 am

Re: Nonparametric Dependence Measures in Quantitative Trading

January 11th, 2018, 1:09 pm

I agree with T$A. I really, honestly don't understand why probability of losses is so high up staassis' list, the measure erases almost all distribution info w.r.t. expected return, risk, risk/return ratios.. all is discarded except a single bit per sample.

..and a per-trade measure means that any dynamic allocation algorithm that everyone use (like Kelly strategies) and which generates a lot of rebalancing trades will make this measure very noisy. And spread trades, ...or dynamic hedging non-linear instruments.. all generate lots of trades who's individual P&L signs aren't very meaningful in isolation.
 
User avatar
staassis
Topic Author
Posts: 27
Joined: April 12th, 2014, 5:10 pm

Re: Nonparametric Dependence Measures in Quantitative Trading

January 11th, 2018, 3:04 pm

Traden4Alpha, trader's know-how is composed of at least 3 segments: 

1) trading signals (detection of new prop opportunities), 
2) execution strategy (entry / exit levels, maximum duration, stop loss rules if any, etc),
3) risk management.

You have mentioned segment 2, where return magnitudes matter a lot. Among other things, the trader needs to make sure that he does not get out of profitable opportunities too early and is well set to ride the move... Generation of signals (segment 1) is another story. If you think that you see something in the market and you are wrong 50% of the time, this is dangerous. You can improve your Sharpe ratio (even if it's positive already) by pausing and identifying flaws in yourself.   
 
User avatar
staassis
Topic Author
Posts: 27
Joined: April 12th, 2014, 5:10 pm

Re: Nonparametric Dependence Measures in Quantitative Trading

January 11th, 2018, 3:06 pm

In an irrational hope to bring this discussion back to the theme of the thread, here's another resource on Spearman's rho. Mann-Kendall and Spearman's rho tests are used for trend detection.

https://rmgsc.cr.usgs.gov/outgoing/thre ... al2002.pdf
 
User avatar
Traden4Alpha
Posts: 3300
Joined: September 20th, 2002, 8:30 pm

Re: Nonparametric Dependence Measures in Quantitative Trading

January 11th, 2018, 10:14 pm

The reason we keep going on about the folly of win-loss ratios is that (besides being dangerously wrong) it only works if all prop traders or trading systems have the same disribution of returns and that distribution is symmetric.  If the distribution is assymetric, the break-even win-loss ratio might be extremely high (e.g., a 90% win rate results in bankruptcy) or extremely low (e.g., a 10% win rate leads to riches).

If you want to do more with non-parametric measures of trading performance, then don't use win-loss ratio.
 
User avatar
staassis
Topic Author
Posts: 27
Joined: April 12th, 2014, 5:10 pm

Re: Nonparametric Dependence Measures in Quantitative Trading

January 12th, 2018, 11:27 am

The reason we keep going on about the folly of win-loss ratios is that (besides being dangerously wrong) it only works if all prop traders or trading systems have the same disribution of returns and that distribution is symmetric.  If the distribution is assymetric, the break-even win-loss ratio might be extremely high (e.g., a 90% win rate results in bankruptcy) or extremely low (e.g., a 10% win rate leads to riches).
You seem to be ignoring my message to Outrun on January 9: "Potential asymmetry of portfolio returns is a well-documented issue. So one needs to make sure that the expected returns and Sharpe ratio are positive and large, as much as live testing and backtesting allow. However, this is not enough..."

I may be wrong, I hope I am, but you sound like you have just put down an intermediate-level probability or risk management book and want to share it with the world. Do you trade? What is your trading performance? More importantly, were you able to consistently make money on a strategy where "10% win rate leads to riches"? Or is this just an idealization which your mind has build based on abstract probabilistic arguments? And how do you know that it is 10% rate and not 5% rate? Do you really know the joint distribution of all relevant financial variables, today, tomorrow?... One useful thing that Outrun did is he initiated the discussion of reinforcement learning. And reinforcement learning teaches us that oftentimes the underlying distribution is unknown. So one has to learn on the run and constantly adapt the strategies to recent history of actions and rewards... Every trade matters, especially an unsuccessful one, because it may signal structural changes in the market.