[amibroker] Re: AIRAP - fitness function
There is also an article on your reference site on MaxDd http://www.intelligenthedgefundinvesting.com/ch06.html. In fact, there were other performance measures for you to consider. while I'll need sometime to look at the AIRAP more closely. I think it is important to point out that it is just as important to understand how different ways of applying your data to generate statistical fitness can affect your final answer just as much as choosing different fitness functions. For example, consider the follwoing Rate of return calculation: Period :Discrete Rate Continuously compounded 1 100% 0.693 =log(1+1) 2 -50% -0.693 =log(1-0.5) Avg 0.25 0 As we can see the arithmetic average, or the mean of the discrete rates of return, is plus 25% per period. Yet the investment has simply doubled and then halved to return to its original value at time 0. --- In amibroker@yahoogroups.com, tf28373 tom...@... wrote: Hello everyone I have been working on the choose of fitness function following the Howard Bundy's advices in his Quantitative Trading Systems and come across M. Sharma's Alternative Investments Risk Adjusted Performance (AIRAP). The equation of it is as following: AIRAP = [ E pi*(1+TRi)(1-c) ] 1/(1-c) - 1, where TRi - ith period total fund return (in my opinon it can also be ith trade net return), c - risk aversion parameter (author suggests to set its value to c=4), i=1,...,N - number of periods (as for me it can be number of trades), pi - the probability of the ith period's total return (according to the author it can be replaced with 1/N). (For futher information please check this working paper: http://www.intelligenthedgefundinvesting.com/pubs/rb-ms01.pdf http://www.intelligenthedgefundinvesting.com/pubs/rb-ms01.pdf .) M. Sharma argues that this measure captures all higher moments, penalizes for higher volatility and leverage (downside risk is penalized more) and has all merits of Sharp ratio, though without its limitations and disadvantages. I have carried out some simulations on the artificial returns of different distributions and indeed it makes some difference. Nevertheless what I am suspicious about is the fact that it was the very first time I found this objective function even though it was created by Sharma about 5 years ago. As for me it can mean that AIRAP is in fact far from being effective or/and practical fitness measure at least for trader like us and nobody use it (maybe I am wrong...). Another issue that concerns me a bit is omission of MaxDrawDown in the equation, which - at least for me - is a very important risk measure. According to many experienced wise people writing on this forum (like ex Mr Bundy), an effective fitness function shouls take Max DD or some comparable risk measure into consideration in order to be really useful. What do you think about AIRAP? Should I proceed with utilizing this function? I am looking forward to your response. Thank you in advance. Tomasz
[amibroker] Re: AIRAP - fitness function
Thank you Paul for your contribution to this discussion. This example you have drawn is very valuable - to be honest until now I have been using discrete rates in all my calculations, which now seems to me as very tricky. Since - as far as I am concerned - AB calculates all performance metrics basing on arithmetic values, it would be advisable to prepere my own measures through CBT. --- In amibroker@yahoogroups.com, paultsho paul.t...@... wrote: There is also an article on your reference site on MaxDd http://www.intelligenthedgefundinvesting.com/ch06.html. In fact, there were other performance measures for you to consider. while I'll need sometime to look at the AIRAP more closely. I think it is important to point out that it is just as important to understand how different ways of applying your data to generate statistical fitness can affect your final answer just as much as choosing different fitness functions. For example, consider the follwoing Rate of return calculation: Period :Discrete Rate Continuously compounded 1 100% 0.693 =log(1+1) 2 -50% -0.693 =log(1-0.5) Avg 0.25 0 As we can see the arithmetic average, or the mean of the discrete rates of return, is plus 25% per period. Yet the investment has simply doubled and then halved to return to its original value at time 0. --- In amibroker@yahoogroups.com, tf28373 tomfid@ wrote: Hello everyone I have been working on the choose of fitness function following the Howard Bundy's advices in his Quantitative Trading Systems and come across M. Sharma's Alternative Investments Risk Adjusted Performance (AIRAP). The equation of it is as following: AIRAP = [ E pi*(1+TRi)(1-c) ] 1/(1-c) - 1, where TRi - ith period total fund return (in my opinon it can also be ith trade net return), c - risk aversion parameter (author suggests to set its value to c=4), i=1,...,N - number of periods (as for me it can be number of trades), pi - the probability of the ith period's total return (according to the author it can be replaced with 1/N). (For futher information please check this working paper: http://www.intelligenthedgefundinvesting.com/pubs/rb-ms01.pdf http://www.intelligenthedgefundinvesting.com/pubs/rb-ms01.pdf .) M. Sharma argues that this measure captures all higher moments, penalizes for higher volatility and leverage (downside risk is penalized more) and has all merits of Sharp ratio, though without its limitations and disadvantages. I have carried out some simulations on the artificial returns of different distributions and indeed it makes some difference. Nevertheless what I am suspicious about is the fact that it was the very first time I found this objective function even though it was created by Sharma about 5 years ago. As for me it can mean that AIRAP is in fact far from being effective or/and practical fitness measure at least for trader like us and nobody use it (maybe I am wrong...). Another issue that concerns me a bit is omission of MaxDrawDown in the equation, which - at least for me - is a very important risk measure. According to many experienced wise people writing on this forum (like ex Mr Bundy), an effective fitness function shouls take Max DD or some comparable risk measure into consideration in order to be really useful. What do you think about AIRAP? Should I proceed with utilizing this function? I am looking forward to your response. Thank you in advance. Tomasz
Re: [amibroker] Re: AIRAP - fitness function
Hi Scott, What about replacing MaxDD by AvgDD + 2 * StdevDD ? Regards, Ton. - Original Message - From: sdwcyberdude To: amibroker@yahoogroups.com Sent: Wednesday, September 08, 2010 4:34 PM Subject: [amibroker] Re: AIRAP - fitness function Tomasz, I also use and see value in the Max DD, however, I believe it is should only be a secondary measure. Think of a 10 year backtest. System X has 1 drawdown of 30% (max), and many small drawdown never exceeding 5%. System Y has 1 drawdown of 25% (max), and 10+ other drawdowns between 20 and 23%. Which system is more stable? I will invest risk capital in System X, which has the higher max drawdown, but much fewer drawdowns of depth. I would love to have a measure of drawdown that more directly and intuitively measures the depth and frequency of drawdowns per unit of time. Correlation of the equity curve also gets at that point. Regarding the Omega, I am relying on a friend who studied both the advanced math and models and uncover significant concerns with the Omega (I seem to recall in was bias issues around skewness and kurtosis, but I might be wrong), however it was unpublished work for hedge funds. He also developed a proprietary alternative. Sorry I can't be more helpful on that one. Kind Regards, Scott --- In amibroker@yahoogroups.com, tf28373 tom...@... wrote: Hi Scott Thanks for response. I agree that the Sortino ratio is a kind of solution to the typical Sharp ratio disadvantages (like penalization high moments, which for me is irrational). Nevertheless, there is no max dd taken into account, which confuses me a bit. However, I might be too devoted to this risk measure (max dd) - what do you think? Is mean and its variance better/sufficient values as far as the characteristics of equity line is considered? (This is what brain123 was supporting in many discussions.) One should be careful if it is built upon the Omega, which I believe introduces other problems. That is an interesting point - can you elaborate a bit on this one? In fact I was hoping to get this kind of information when starting this thread as - frankly speaking - I don't feel familiar with plain maths enough to analyse it... Looking forward to your response. Regards Tomasz --- In amibroker@yahoogroups.com, sdwcyberdude scwalker1986@ wrote: Tomasz, Thanks for raising this question (and for the good work you do). The Sortino Ratio is a well regarding improvement upon the Sharpe; I urge you to consider adding the Sortino to the base metric array. Is there a reason you passed on it earlier? The Sharpe ratio has a lot of problems and I was not familiar with the AIRAP. One should be careful if it is built upon the Omega, which I believe introduces other problems. Regards, Scott --- In amibroker@yahoogroups.com, tf28373 tomfid@ wrote: Hello everyone I have been working on the choose of fitness function following the Howard Bundy's advices in his Quantitative Trading Systems and come across M. Sharma's Alternative Investments Risk Adjusted Performance (AIRAP). The equation of it is as following: AIRAP = [ E pi*(1+TRi)(1-c) ] 1/(1-c) - 1, where TRi - ith period total fund return (in my opinon it can also be ith trade net return), c - risk aversion parameter (author suggests to set its value to c=4), i=1,...,N - number of periods (as for me it can be number of trades), pi - the probability of the ith period's total return (according to the author it can be replaced with 1/N). (For futher information please check this working paper: http://www.intelligenthedgefundinvesting.com/pubs/rb-ms01.pdf http://www.intelligenthedgefundinvesting.com/pubs/rb-ms01.pdf .) M. Sharma argues that this measure captures all higher moments, penalizes for higher volatility and leverage (downside risk is penalized more) and has all merits of Sharp ratio, though without its limitations and disadvantages. I have carried out some simulations on the artificial returns of different distributions and indeed it makes some difference. Nevertheless what I am suspicious about is the fact that it was the very first time I found this objective function even though it was created by Sharma about 5 years ago. As for me it can mean that AIRAP is in fact far from being effective or/and practical fitness measure at least for trader like us and nobody use it (maybe I am wrong...). Another issue that concerns me a bit is omission of MaxDrawDown in the equation, which - at least for me - is a very important risk measure. According to many experienced wise people writing on this forum (like ex Mr Bundy), an effective fitness function shouls take Max DD or some comparable risk
[amibroker] Re: AIRAP - fitness function
Ton, Appears to be the right direction + promising; perhaps one of the forum math mavens will comment. Regards, Scott --- In amibroker@yahoogroups.com, Ton Sieverding ton.sieverd...@... wrote: Hi Scott, What about replacing MaxDD by AvgDD + 2 * StdevDD ? Regards, Ton. - Original Message - From: sdwcyberdude To: amibroker@yahoogroups.com Sent: Wednesday, September 08, 2010 4:34 PM Subject: [amibroker] Re: AIRAP - fitness function Tomasz, I also use and see value in the Max DD, however, I believe it is should only be a secondary measure. Think of a 10 year backtest. System X has 1 drawdown of 30% (max), and many small drawdown never exceeding 5%. System Y has 1 drawdown of 25% (max), and 10+ other drawdowns between 20 and 23%. Which system is more stable? I will invest risk capital in System X, which has the higher max drawdown, but much fewer drawdowns of depth. I would love to have a measure of drawdown that more directly and intuitively measures the depth and frequency of drawdowns per unit of time. Correlation of the equity curve also gets at that point. Regarding the Omega, I am relying on a friend who studied both the advanced math and models and uncover significant concerns with the Omega (I seem to recall in was bias issues around skewness and kurtosis, but I might be wrong), however it was unpublished work for hedge funds. He also developed a proprietary alternative. Sorry I can't be more helpful on that one. Kind Regards, Scott --- In amibroker@yahoogroups.com, tf28373 tomfid@ wrote: Hi Scott Thanks for response. I agree that the Sortino ratio is a kind of solution to the typical Sharp ratio disadvantages (like penalization high moments, which for me is irrational). Nevertheless, there is no max dd taken into account, which confuses me a bit. However, I might be too devoted to this risk measure (max dd) - what do you think? Is mean and its variance better/sufficient values as far as the characteristics of equity line is considered? (This is what brain123 was supporting in many discussions.) One should be careful if it is built upon the Omega, which I believe introduces other problems. That is an interesting point - can you elaborate a bit on this one? In fact I was hoping to get this kind of information when starting this thread as - frankly speaking - I don't feel familiar with plain maths enough to analyse it... Looking forward to your response. Regards Tomasz --- In amibroker@yahoogroups.com, sdwcyberdude scwalker1986@ wrote: Tomasz, Thanks for raising this question (and for the good work you do). The Sortino Ratio is a well regarding improvement upon the Sharpe; I urge you to consider adding the Sortino to the base metric array. Is there a reason you passed on it earlier? The Sharpe ratio has a lot of problems and I was not familiar with the AIRAP. One should be careful if it is built upon the Omega, which I believe introduces other problems. Regards, Scott --- In amibroker@yahoogroups.com, tf28373 tomfid@ wrote: Hello everyone I have been working on the choose of fitness function following the Howard Bundy's advices in his Quantitative Trading Systems and come across M. Sharma's Alternative Investments Risk Adjusted Performance (AIRAP). The equation of it is as following: AIRAP = [ E pi*(1+TRi)(1-c) ] 1/(1-c) - 1, where TRi - ith period total fund return (in my opinon it can also be ith trade net return), c - risk aversion parameter (author suggests to set its value to c=4), i=1,...,N - number of periods (as for me it can be number of trades), pi - the probability of the ith period's total return (according to the author it can be replaced with 1/N). (For futher information please check this working paper: http://www.intelligenthedgefundinvesting.com/pubs/rb-ms01.pdf http://www.intelligenthedgefundinvesting.com/pubs/rb-ms01.pdf .) M. Sharma argues that this measure captures all higher moments, penalizes for higher volatility and leverage (downside risk is penalized more) and has all merits of Sharp ratio, though without its limitations and disadvantages. I have carried out some simulations on the artificial returns of different distributions and indeed it makes some difference. Nevertheless what I am suspicious about is the fact that it was the very first time I found this objective function even though it was created by Sharma about 5 years ago. As for me it can mean that AIRAP is in fact far from being effective or/and practical fitness measure at least for trader like us and nobody use it (maybe I am wrong...). Another
[amibroker] Re: AIRAP - fitness function
Tomasz, I also use and see value in the Max DD, however, I believe it is should only be a secondary measure. Think of a 10 year backtest. System X has 1 drawdown of 30% (max), and many small drawdown never exceeding 5%. System Y has 1 drawdown of 25% (max), and 10+ other drawdowns between 20 and 23%. Which system is more stable? I will invest risk capital in System X, which has the higher max drawdown, but much fewer drawdowns of depth. I would love to have a measure of drawdown that more directly and intuitively measures the depth and frequency of drawdowns per unit of time. Correlation of the equity curve also gets at that point. Regarding the Omega, I am relying on a friend who studied both the advanced math and models and uncover significant concerns with the Omega (I seem to recall in was bias issues around skewness and kurtosis, but I might be wrong), however it was unpublished work for hedge funds. He also developed a proprietary alternative. Sorry I can't be more helpful on that one. Kind Regards, Scott --- In amibroker@yahoogroups.com, tf28373 tom...@... wrote: Hi Scott Thanks for response. I agree that the Sortino ratio is a kind of solution to the typical Sharp ratio disadvantages (like penalization high moments, which for me is irrational). Nevertheless, there is no max dd taken into account, which confuses me a bit. However, I might be too devoted to this risk measure (max dd) - what do you think? Is mean and its variance better/sufficient values as far as the characteristics of equity line is considered? (This is what brain123 was supporting in many discussions.) One should be careful if it is built upon the Omega, which I believe introduces other problems. That is an interesting point - can you elaborate a bit on this one? In fact I was hoping to get this kind of information when starting this thread as - frankly speaking - I don't feel familiar with plain maths enough to analyse it... Looking forward to your response. Regards Tomasz --- In amibroker@yahoogroups.com, sdwcyberdude scwalker1986@ wrote: Tomasz, Thanks for raising this question (and for the good work you do). The Sortino Ratio is a well regarding improvement upon the Sharpe; I urge you to consider adding the Sortino to the base metric array. Is there a reason you passed on it earlier? The Sharpe ratio has a lot of problems and I was not familiar with the AIRAP. One should be careful if it is built upon the Omega, which I believe introduces other problems. Regards, Scott --- In amibroker@yahoogroups.com, tf28373 tomfid@ wrote: Hello everyone I have been working on the choose of fitness function following the Howard Bundy's advices in his Quantitative Trading Systems and come across M. Sharma's Alternative Investments Risk Adjusted Performance (AIRAP). The equation of it is as following: AIRAP = [ E pi*(1+TRi)(1-c) ] 1/(1-c) - 1, where TRi - ith period total fund return (in my opinon it can also be ith trade net return), c - risk aversion parameter (author suggests to set its value to c=4), i=1,...,N - number of periods (as for me it can be number of trades), pi - the probability of the ith period's total return (according to the author it can be replaced with 1/N). (For futher information please check this working paper: http://www.intelligenthedgefundinvesting.com/pubs/rb-ms01.pdf http://www.intelligenthedgefundinvesting.com/pubs/rb-ms01.pdf .) M. Sharma argues that this measure captures all higher moments, penalizes for higher volatility and leverage (downside risk is penalized more) and has all merits of Sharp ratio, though without its limitations and disadvantages. I have carried out some simulations on the artificial returns of different distributions and indeed it makes some difference. Nevertheless what I am suspicious about is the fact that it was the very first time I found this objective function even though it was created by Sharma about 5 years ago. As for me it can mean that AIRAP is in fact far from being effective or/and practical fitness measure at least for trader like us and nobody use it (maybe I am wrong...). Another issue that concerns me a bit is omission of MaxDrawDown in the equation, which - at least for me - is a very important risk measure. According to many experienced wise people writing on this forum (like ex Mr Bundy), an effective fitness function shouls take Max DD or some comparable risk measure into consideration in order to be really useful. What do you think about AIRAP? Should I proceed with utilizing this function? I am looking forward to your response. Thank you in advance. Tomasz
[amibroker] Re: AIRAP - fitness function
I strongly agree. This is the moment when Ulcer Performance Index steps in - check this one out, since it is based on measuring risk in the terms of drawdown regarding its duration. By the way thanks for your comments about Omega - it seems that on the one hand AIRAP can add something positive into system performance analysis, on the other - it has some drawbacks. Anyway my mind is still overoccupied by the idea of deriving the system virtues and equity line features from simple mean - deviation analysis (together with using such indicators like profit factor, power factor, expectation). Have you ever ponder on this one? --- In amibroker@yahoogroups.com, sdwcyberdude scwalker1...@... wrote: Tomasz, I also use and see value in the Max DD, however, I believe it is should only be a secondary measure. Think of a 10 year backtest. System X has 1 drawdown of 30% (max), and many small drawdown never exceeding 5%. System Y has 1 drawdown of 25% (max), and 10+ other drawdowns between 20 and 23%. Which system is more stable? I will invest risk capital in System X, which has the higher max drawdown, but much fewer drawdowns of depth. I would love to have a measure of drawdown that more directly and intuitively measures the depth and frequency of drawdowns per unit of time. Correlation of the equity curve also gets at that point. Regarding the Omega, I am relying on a friend who studied both the advanced math and models and uncover significant concerns with the Omega (I seem to recall in was bias issues around skewness and kurtosis, but I might be wrong), however it was unpublished work for hedge funds. He also developed a proprietary alternative. Sorry I can't be more helpful on that one. Kind Regards, Scott --- In amibroker@yahoogroups.com, tf28373 tomfid@ wrote: Hi Scott Thanks for response. I agree that the Sortino ratio is a kind of solution to the typical Sharp ratio disadvantages (like penalization high moments, which for me is irrational). Nevertheless, there is no max dd taken into account, which confuses me a bit. However, I might be too devoted to this risk measure (max dd) - what do you think? Is mean and its variance better/sufficient values as far as the characteristics of equity line is considered? (This is what brain123 was supporting in many discussions.) One should be careful if it is built upon the Omega, which I believe introduces other problems. That is an interesting point - can you elaborate a bit on this one? In fact I was hoping to get this kind of information when starting this thread as - frankly speaking - I don't feel familiar with plain maths enough to analyse it... Looking forward to your response. Regards Tomasz --- In amibroker@yahoogroups.com, sdwcyberdude scwalker1986@ wrote: Tomasz, Thanks for raising this question (and for the good work you do). The Sortino Ratio is a well regarding improvement upon the Sharpe; I urge you to consider adding the Sortino to the base metric array. Is there a reason you passed on it earlier? The Sharpe ratio has a lot of problems and I was not familiar with the AIRAP. One should be careful if it is built upon the Omega, which I believe introduces other problems. Regards, Scott --- In amibroker@yahoogroups.com, tf28373 tomfid@ wrote: Hello everyone I have been working on the choose of fitness function following the Howard Bundy's advices in his Quantitative Trading Systems and come across M. Sharma's Alternative Investments Risk Adjusted Performance (AIRAP). The equation of it is as following: AIRAP = [ E pi*(1+TRi)(1-c) ] 1/(1-c) - 1, where TRi - ith period total fund return (in my opinon it can also be ith trade net return), c - risk aversion parameter (author suggests to set its value to c=4), i=1,...,N - number of periods (as for me it can be number of trades), pi - the probability of the ith period's total return (according to the author it can be replaced with 1/N). (For futher information please check this working paper: http://www.intelligenthedgefundinvesting.com/pubs/rb-ms01.pdf http://www.intelligenthedgefundinvesting.com/pubs/rb-ms01.pdf .) M. Sharma argues that this measure captures all higher moments, penalizes for higher volatility and leverage (downside risk is penalized more) and has all merits of Sharp ratio, though without its limitations and disadvantages. I have carried out some simulations on the artificial returns of different distributions and indeed it makes some difference. Nevertheless what I am suspicious about is the fact that it was the very first time I found this objective function even though it was created by Sharma about 5 years ago. As for me it can mean that AIRAP is in fact far from being effective
[amibroker] Re: AIRAP - fitness function
Good points. I strive to spend 90% of my focus on deriving and measuring the underlying edge, why does it exist, when did it exist does it still exist, how is it captured + measured. Last 10% is spent on the system metrics, etc. The big dog, The Edge, seems not to be discussed very much by many people. Regards, Scott --- In amibroker@yahoogroups.com, tf28373 tom...@... wrote: I strongly agree. This is the moment when Ulcer Performance Index steps in - check this one out, since it is based on measuring risk in the terms of drawdown regarding its duration. By the way thanks for your comments about Omega - it seems that on the one hand AIRAP can add something positive into system performance analysis, on the other - it has some drawbacks. Anyway my mind is still overoccupied by the idea of deriving the system virtues and equity line features from simple mean - deviation analysis (together with using such indicators like profit factor, power factor, expectation). Have you ever ponder on this one? --- In amibroker@yahoogroups.com, sdwcyberdude scwalker1986@ wrote: Tomasz, I also use and see value in the Max DD, however, I believe it is should only be a secondary measure. Think of a 10 year backtest. System X has 1 drawdown of 30% (max), and many small drawdown never exceeding 5%. System Y has 1 drawdown of 25% (max), and 10+ other drawdowns between 20 and 23%. Which system is more stable? I will invest risk capital in System X, which has the higher max drawdown, but much fewer drawdowns of depth. I would love to have a measure of drawdown that more directly and intuitively measures the depth and frequency of drawdowns per unit of time. Correlation of the equity curve also gets at that point. Regarding the Omega, I am relying on a friend who studied both the advanced math and models and uncover significant concerns with the Omega (I seem to recall in was bias issues around skewness and kurtosis, but I might be wrong), however it was unpublished work for hedge funds. He also developed a proprietary alternative. Sorry I can't be more helpful on that one. Kind Regards, Scott --- In amibroker@yahoogroups.com, tf28373 tomfid@ wrote: Hi Scott Thanks for response. I agree that the Sortino ratio is a kind of solution to the typical Sharp ratio disadvantages (like penalization high moments, which for me is irrational). Nevertheless, there is no max dd taken into account, which confuses me a bit. However, I might be too devoted to this risk measure (max dd) - what do you think? Is mean and its variance better/sufficient values as far as the characteristics of equity line is considered? (This is what brain123 was supporting in many discussions.) One should be careful if it is built upon the Omega, which I believe introduces other problems. That is an interesting point - can you elaborate a bit on this one? In fact I was hoping to get this kind of information when starting this thread as - frankly speaking - I don't feel familiar with plain maths enough to analyse it... Looking forward to your response. Regards Tomasz --- In amibroker@yahoogroups.com, sdwcyberdude scwalker1986@ wrote: Tomasz, Thanks for raising this question (and for the good work you do). The Sortino Ratio is a well regarding improvement upon the Sharpe; I urge you to consider adding the Sortino to the base metric array. Is there a reason you passed on it earlier? The Sharpe ratio has a lot of problems and I was not familiar with the AIRAP. One should be careful if it is built upon the Omega, which I believe introduces other problems. Regards, Scott --- In amibroker@yahoogroups.com, tf28373 tomfid@ wrote: Hello everyone I have been working on the choose of fitness function following the Howard Bundy's advices in his Quantitative Trading Systems and come across M. Sharma's Alternative Investments Risk Adjusted Performance (AIRAP). The equation of it is as following: AIRAP = [ E pi*(1+TRi)(1-c) ] 1/(1-c) - 1, where TRi - ith period total fund return (in my opinon it can also be ith trade net return), c - risk aversion parameter (author suggests to set its value to c=4), i=1,...,N - number of periods (as for me it can be number of trades), pi - the probability of the ith period's total return (according to the author it can be replaced with 1/N). (For futher information please check this working paper: http://www.intelligenthedgefundinvesting.com/pubs/rb-ms01.pdf http://www.intelligenthedgefundinvesting.com/pubs/rb-ms01.pdf .) M. Sharma argues that this measure captures all higher moments, penalizes for higher volatility and leverage (downside risk
Re: [amibroker] Re: AIRAP - fitness function
Hi Scott -- If by edge you mean expectancy, then it is well understood as being very important and often discussed in this forum. Or do you have a different, and quantifiable, definition of edge? Thanks, Howard On Wed, Sep 8, 2010 at 9:48 AM, sdwcyberdude scwalker1...@gmail.com wrote: Good points. I strive to spend 90% of my focus on deriving and measuring the underlying edge, why does it exist, when did it exist does it still exist, how is it captured + measured. Last 10% is spent on the system metrics, etc. The big dog, The Edge, seems not to be discussed very much by many people. Regards, Scott --- In amibroker@yahoogroups.com amibroker%40yahoogroups.com, tf28373 tom...@... wrote: I strongly agree. This is the moment when Ulcer Performance Index steps in - check this one out, since it is based on measuring risk in the terms of drawdown regarding its duration. By the way thanks for your comments about Omega - it seems that on the one hand AIRAP can add something positive into system performance analysis, on the other - it has some drawbacks. Anyway my mind is still overoccupied by the idea of deriving the system virtues and equity line features from simple mean - deviation analysis (together with using such indicators like profit factor, power factor, expectation). Have you ever ponder on this one? --- In amibroker@yahoogroups.com amibroker%40yahoogroups.com, sdwcyberdude scwalker1986@ wrote: Tomasz, I also use and see value in the Max DD, however, I believe it is should only be a secondary measure. Think of a 10 year backtest. System X has 1 drawdown of 30% (max), and many small drawdown never exceeding 5%. System Y has 1 drawdown of 25% (max), and 10+ other drawdowns between 20 and 23%. Which system is more stable? I will invest risk capital in System X, which has the higher max drawdown, but much fewer drawdowns of depth. I would love to have a measure of drawdown that more directly and intuitively measures the depth and frequency of drawdowns per unit of time. Correlation of the equity curve also gets at that point. Regarding the Omega, I am relying on a friend who studied both the advanced math and models and uncover significant concerns with the Omega (I seem to recall in was bias issues around skewness and kurtosis, but I might be wrong), however it was unpublished work for hedge funds. He also developed a proprietary alternative. Sorry I can't be more helpful on that one. Kind Regards, Scott --- In amibroker@yahoogroups.com amibroker%40yahoogroups.com, tf28373 tomfid@ wrote: Hi Scott Thanks for response. I agree that the Sortino ratio is a kind of solution to the typical Sharp ratio disadvantages (like penalization high moments, which for me is irrational). Nevertheless, there is no max dd taken into account, which confuses me a bit. However, I might be too devoted to this risk measure (max dd) - what do you think? Is mean and its variance better/sufficient values as far as the characteristics of equity line is considered? (This is what brain123 was supporting in many discussions.) One should be careful if it is built upon the Omega, which I believe introduces other problems. That is an interesting point - can you elaborate a bit on this one? In fact I was hoping to get this kind of information when starting this thread as - frankly speaking - I don't feel familiar with plain maths enough to analyse it... Looking forward to your response. Regards Tomasz --- In amibroker@yahoogroups.com amibroker%40yahoogroups.com, sdwcyberdude scwalker1986@ wrote: Tomasz, Thanks for raising this question (and for the good work you do). The Sortino Ratio is a well regarding improvement upon the Sharpe; I urge you to consider adding the Sortino to the base metric array. Is there a reason you passed on it earlier? The Sharpe ratio has a lot of problems and I was not familiar with the AIRAP. One should be careful if it is built upon the Omega, which I believe introduces other problems. Regards, Scott --- In amibroker@yahoogroups.com amibroker%40yahoogroups.com, tf28373 tomfid@ wrote: Hello everyone I have been working on the choose of fitness function following the Howard Bundy's advices in his Quantitative Trading Systems and come across M. Sharma's Alternative Investments Risk Adjusted Performance (AIRAP). The equation of it is as following: AIRAP = [ E pi*(1+TRi)(1-c) ] 1/(1-c) - 1, where TRi - ith period total fund return (in my opinon it can also be ith trade net return), c - risk aversion parameter (author suggests to set its value to c=4), i=1,...,N - number of periods (as for me it can be number of trades), pi - the probability of the ith period's total return
[amibroker] Re: AIRAP - fitness function
If by expectancy you mean something along the lines of average net profit/trade, then no. By the edge I refer to underlying concepts (e.g., the shifting relationship of mean reversion vs. trend following with different asset classes/market regimes). A system measurement (like expectancy) is only useful and important after all work of - capturing the truth, and - determining how/why it works and if it is continuing, and - under what circumstances would it be expected to continue and - translating this into a process/system edge (competitive advantage) in trading/management. Regards, Scott --- In amibroker@yahoogroups.com, Howard B howardba...@... wrote: Hi Scott -- If by edge you mean expectancy, then it is well understood as being very important and often discussed in this forum. Or do you have a different, and quantifiable, definition of edge? Thanks, Howard On Wed, Sep 8, 2010 at 9:48 AM, sdwcyberdude scwalker1...@... wrote: Good points. I strive to spend 90% of my focus on deriving and measuring the underlying edge, why does it exist, when did it exist does it still exist, how is it captured + measured. Last 10% is spent on the system metrics, etc. The big dog, The Edge, seems not to be discussed very much by many people. Regards, Scott --- In amibroker@yahoogroups.com amibroker%40yahoogroups.com, tf28373 tomfid@ wrote: I strongly agree. This is the moment when Ulcer Performance Index steps in - check this one out, since it is based on measuring risk in the terms of drawdown regarding its duration. By the way thanks for your comments about Omega - it seems that on the one hand AIRAP can add something positive into system performance analysis, on the other - it has some drawbacks. Anyway my mind is still overoccupied by the idea of deriving the system virtues and equity line features from simple mean - deviation analysis (together with using such indicators like profit factor, power factor, expectation). Have you ever ponder on this one? --- In amibroker@yahoogroups.com amibroker%40yahoogroups.com, sdwcyberdude scwalker1986@ wrote: Tomasz, I also use and see value in the Max DD, however, I believe it is should only be a secondary measure. Think of a 10 year backtest. System X has 1 drawdown of 30% (max), and many small drawdown never exceeding 5%. System Y has 1 drawdown of 25% (max), and 10+ other drawdowns between 20 and 23%. Which system is more stable? I will invest risk capital in System X, which has the higher max drawdown, but much fewer drawdowns of depth. I would love to have a measure of drawdown that more directly and intuitively measures the depth and frequency of drawdowns per unit of time. Correlation of the equity curve also gets at that point. Regarding the Omega, I am relying on a friend who studied both the advanced math and models and uncover significant concerns with the Omega (I seem to recall in was bias issues around skewness and kurtosis, but I might be wrong), however it was unpublished work for hedge funds. He also developed a proprietary alternative. Sorry I can't be more helpful on that one. Kind Regards, Scott --- In amibroker@yahoogroups.com amibroker%40yahoogroups.com, tf28373 tomfid@ wrote: Hi Scott Thanks for response. I agree that the Sortino ratio is a kind of solution to the typical Sharp ratio disadvantages (like penalization high moments, which for me is irrational). Nevertheless, there is no max dd taken into account, which confuses me a bit. However, I might be too devoted to this risk measure (max dd) - what do you think? Is mean and its variance better/sufficient values as far as the characteristics of equity line is considered? (This is what brain123 was supporting in many discussions.) One should be careful if it is built upon the Omega, which I believe introduces other problems. That is an interesting point - can you elaborate a bit on this one? In fact I was hoping to get this kind of information when starting this thread as - frankly speaking - I don't feel familiar with plain maths enough to analyse it... Looking forward to your response. Regards Tomasz --- In amibroker@yahoogroups.com amibroker%40yahoogroups.com, sdwcyberdude scwalker1986@ wrote: Tomasz, Thanks for raising this question (and for the good work you do). The Sortino Ratio is a well regarding improvement upon the Sharpe; I urge you to consider adding the Sortino to the base metric array. Is there a reason you passed on it earlier? The Sharpe ratio has a lot of problems and I was not familiar with the AIRAP. One should be careful if it is built upon the Omega, which I believe introduces other problems. Regards, Scott --- In
[amibroker] Re: AIRAP - fitness function
Tomasz, Thanks for raising this question (and for the good work you do). The Sortino Ratio is a well regarding improvement upon the Sharpe; I urge you to consider adding the Sortino to the base metric array. Is there a reason you passed on it earlier? The Sharpe ratio has a lot of problems and I was not familiar with the AIRAP. One should be careful if it is built upon the Omega, which I believe introduces other problems. Regards, Scott --- In amibroker@yahoogroups.com, tf28373 tom...@... wrote: Hello everyone I have been working on the choose of fitness function following the Howard Bundy's advices in his Quantitative Trading Systems and come across M. Sharma's Alternative Investments Risk Adjusted Performance (AIRAP). The equation of it is as following: AIRAP = [ E pi*(1+TRi)(1-c) ] 1/(1-c) - 1, where TRi - ith period total fund return (in my opinon it can also be ith trade net return), c - risk aversion parameter (author suggests to set its value to c=4), i=1,...,N - number of periods (as for me it can be number of trades), pi - the probability of the ith period's total return (according to the author it can be replaced with 1/N). (For futher information please check this working paper: http://www.intelligenthedgefundinvesting.com/pubs/rb-ms01.pdf http://www.intelligenthedgefundinvesting.com/pubs/rb-ms01.pdf .) M. Sharma argues that this measure captures all higher moments, penalizes for higher volatility and leverage (downside risk is penalized more) and has all merits of Sharp ratio, though without its limitations and disadvantages. I have carried out some simulations on the artificial returns of different distributions and indeed it makes some difference. Nevertheless what I am suspicious about is the fact that it was the very first time I found this objective function even though it was created by Sharma about 5 years ago. As for me it can mean that AIRAP is in fact far from being effective or/and practical fitness measure at least for trader like us and nobody use it (maybe I am wrong...). Another issue that concerns me a bit is omission of MaxDrawDown in the equation, which - at least for me - is a very important risk measure. According to many experienced wise people writing on this forum (like ex Mr Bundy), an effective fitness function shouls take Max DD or some comparable risk measure into consideration in order to be really useful. What do you think about AIRAP? Should I proceed with utilizing this function? I am looking forward to your response. Thank you in advance. Tomasz
[amibroker] Re: AIRAP - fitness function
Hi Scott Thanks for response. I agree that the Sortino ratio is a kind of solution to the typical Sharp ratio disadvantages (like penalization high moments, which for me is irrational). Nevertheless, there is no max dd taken into account, which confuses me a bit. However, I might be too devoted to this risk measure (max dd) - what do you think? Is mean and its variance better/sufficient values as far as the characteristics of equity line is considered? (This is what brain123 was supporting in many discussions.) One should be careful if it is built upon the Omega, which I believe introduces other problems. That is an interesting point - can you elaborate a bit on this one? In fact I was hoping to get this kind of information when starting this thread as - frankly speaking - I don't feel familiar with plain maths enough to analyse it... Looking forward to your response. Regards Tomasz --- In amibroker@yahoogroups.com, sdwcyberdude scwalker1...@... wrote: Tomasz, Thanks for raising this question (and for the good work you do). The Sortino Ratio is a well regarding improvement upon the Sharpe; I urge you to consider adding the Sortino to the base metric array. Is there a reason you passed on it earlier? The Sharpe ratio has a lot of problems and I was not familiar with the AIRAP. One should be careful if it is built upon the Omega, which I believe introduces other problems. Regards, Scott --- In amibroker@yahoogroups.com, tf28373 tomfid@ wrote: Hello everyone I have been working on the choose of fitness function following the Howard Bundy's advices in his Quantitative Trading Systems and come across M. Sharma's Alternative Investments Risk Adjusted Performance (AIRAP). The equation of it is as following: AIRAP = [ E pi*(1+TRi)(1-c) ] 1/(1-c) - 1, where TRi - ith period total fund return (in my opinon it can also be ith trade net return), c - risk aversion parameter (author suggests to set its value to c=4), i=1,...,N - number of periods (as for me it can be number of trades), pi - the probability of the ith period's total return (according to the author it can be replaced with 1/N). (For futher information please check this working paper: http://www.intelligenthedgefundinvesting.com/pubs/rb-ms01.pdf http://www.intelligenthedgefundinvesting.com/pubs/rb-ms01.pdf .) M. Sharma argues that this measure captures all higher moments, penalizes for higher volatility and leverage (downside risk is penalized more) and has all merits of Sharp ratio, though without its limitations and disadvantages. I have carried out some simulations on the artificial returns of different distributions and indeed it makes some difference. Nevertheless what I am suspicious about is the fact that it was the very first time I found this objective function even though it was created by Sharma about 5 years ago. As for me it can mean that AIRAP is in fact far from being effective or/and practical fitness measure at least for trader like us and nobody use it (maybe I am wrong...). Another issue that concerns me a bit is omission of MaxDrawDown in the equation, which - at least for me - is a very important risk measure. According to many experienced wise people writing on this forum (like ex Mr Bundy), an effective fitness function shouls take Max DD or some comparable risk measure into consideration in order to be really useful. What do you think about AIRAP? Should I proceed with utilizing this function? I am looking forward to your response. Thank you in advance. Tomasz