Quantitative investment and trading ideas, research, and analysis.
Tuesday, August 14, 2007
CNBC Interview video
Here is the link to my interview on why quantitative models are losing money of late:
9 comments:
Anonymous
said...
Hmm, can you substantiate any of the claims you are making?
1. A "factor" is just a variable in a model. There is no apriori reason why it should limit model's applicability to longer time ranges. For example, if the variable maps to an instrument that is liquid and has a high trading frequency in its own right.
2. Isn't it rather cliche to claim that if a hedge fund had a large loss it was because of "overleveraging"? Do you have real insight in the what happened in Goldman Sach's case (such as the nature of their quantitative model ) or are you just guessing like anyone who reads public news can?
1. A multi-billion-dollar fund cannot respond to high-frequency factors without incurring unacceptable trading costs in the form of market impact. Hence many if not most such factors are of a fundamental nature. Based on this consideration, I would argue that this factor model must have a longer holding period than a few days.
2. No one except the fund managers themselves are privy to proprietary information such as which factor is "failing". External analysts like ourselves can only infer from publicly available information as well as their own past experience. That said, it is often impossible to say whether a low-frequency factor has "failed" in a short time frame, as I have explained elsewhere on this blog, even if you are the fund manager. The nature of such factor models isthat you have to run it for many quarters if not years to see if the factors are any good ex-post. Given this state of uncertainty, a fund manager can only argue that the unacceptable drawdown is due to overleveraging (i.e. underestimation of risk), and not any fundamental defect of the model.
With regard to which factors have failed it is interesting to note that value type factors have turned perverse in a dramatic fashion in August without notable sector biases. The level is such that it is unlikely to be a random event.
This could indicate selling by a few large Hedge fund or Quant shops, since there is a tendency for many to use some of the same factors yet the behavior would imply that they are doing the opposite of what one might expect from their following their models on the long and short side. Could it be that they are forced to sell their equity positions and buy to cover in order to avoid doing so on their less liquid credit or fixed income side of their portfolio. This has pressure on certain types of securities in an unexpected fashion.
Saw some interesting data from Lehman on this theory. Any thoughts?
Indeed, if many models share similar factors, then in times of market stress, all of them will suffer drawdowns simultaneously. Given a drawdown, risk management technique, such as the Kelly's formula that I have discussed in this blog, would dictate an immediate reduction of portfolio size in order to preserve capital ("deleveraging" in popular parlance. But actually, if you follow Kelly's formula, even if we keep the same leverage ratio, the portfolio size will have to be reduced due to the reduction of equity as a result of loss.) Unfortunately, a simultaneous reduction of portfolio in multiple funds amounts to a fire sale, further driving down asset prices. A vicious cycle thus ensues. This was what happened when LTCM collapsed, and this may be happening again now.
1. A multi-billion-dollar fund cannot respond to high-frequency factors without incurring unacceptable trading costs in the form of market impact. Hence many if not most such factors are of a fundamental nature. Based on this consideration, I would argue that this factor model must have a longer holding period than a few days.
Wouldn't RenTech be a rather prominent counter-example? They are responsible for up to 10% of NASDAQ trading on some days.
Furthermore, you are neglecting the possibility whereby market impact is the strategy or the model.
Just because 10% of Nasdaq trading is due to Ren Tec does not mean they are engaging in high frequency trading. Liquidating a vast number of huge positions that have been held for 1 month over the course of a day may generate the same volume.
I fail to see how market impact generated by your own trading can be profitable. Perhaps you can construct an example scenario to illustrate this strategy?
Of course, RenTech are very secretive about how they do what they do. The fact that they are very quantitative model-driven is a well established fact (based on their hiring practices and info disclosed by Jim Simmons himself).
Furthermore, based on references in various interviews with Simmons, high-frequency approach seems to be the consensus of many observers of RenTech. Because probability distributions in the markets are constantly changing, as soon as you've found a stable pattern you need to trade on it. Conversely, a pattern needs to be only stable enough for them to act on it. RenTech can certainly do that, having direct market access at several key exchanges. It all fits.
I fail to see how market impact generated by your own trading can be profitable. Perhaps you can construct an example scenario to illustrate this strategy?
Regarding market impact being part of the strategy, the idea is actually quite old. People can't help being somewhat irrational in their reactions to news and price changes -- the fact that can be exploited as yet another market inefficiency. Recall Jesse Livermore "testing" market depth with probing orders. Quoting Jim Simmons again, "We will do experiments of various sorts ... and in some cases we actually have to move large amounts of money through the system and see what happens. In a given experiment we might pump billions of dollars through the market..."
Now I've given you enough hints to understand how market impact can be a part of a profitable strategy...
Yes, we are well-aware that RenTec is quantitatively driven. I have written elsewhere on this blog about them and some of their possible strategies. In fact, in that article I speculated that they employ high frequency strategies for both stock and futures trading. This is not to say, however, that the loss in their Medallion fund is necessarily due to high frequency strategies, since it is a multi-strategy equity fund. In fact, based on everything I have heard and read so far, this is a highly unlikely scenario.
Regarding using market impact to manipulate prices and create panic in order to profit from that, I believe (based on faint memory of my Series 7 exam) this is a violation of SEC rules. Perhaps readers with sounder legal knowledge can confirm or refute this?
Actually, I found something on the books regarding the illegality of your market impact strategy:
Securities Exchange Act of 1934 Section 9 -- Manipulation of Security Prices a. Transactions relating to purchase or sale of security
It shall be unlawful for any person, directly or indirectly, by the use of the mails or any means or instrumentality of interstate commerce, or of any facility of any national securities exchange, ...
1. ... 2. To effect, alone or with one or more other persons, a series of transactions in any security registered on a national securities exchange ..., ... raising or depressing the price of such security, for the purpose of inducing the purchase or sale of such security by others.
9 comments:
Hmm, can you substantiate any of the claims you are making?
1. A "factor" is just a variable in a model. There is no apriori reason why it should limit model's applicability to longer time ranges. For example, if the variable maps to an instrument that is liquid and has a high trading frequency in its own right.
2. Isn't it rather cliche to claim that if a hedge fund had a large loss it was because of "overleveraging"? Do you have real insight in the what happened in Goldman Sach's case (such as the nature of their quantitative model ) or are you just guessing like anyone who reads public news can?
Dear anonymous
1. A multi-billion-dollar fund cannot respond to high-frequency factors without incurring unacceptable trading costs in the form of market impact. Hence many if not most such factors are of a fundamental nature. Based on this consideration, I would argue that this factor model must have a longer holding period than a few days.
2. No one except the fund managers themselves are privy to proprietary information such as which factor is "failing". External analysts like ourselves can only infer from publicly available information as well as their own past experience. That said, it is often impossible to say whether a low-frequency factor has "failed" in a short time frame, as I have explained elsewhere on this blog, even if you are the fund manager. The nature of such factor models isthat you have to run it for many quarters if not years to see if the factors are any good ex-post. Given this state of uncertainty, a fund manager can only argue that the unacceptable drawdown is due to overleveraging (i.e. underestimation of risk), and not any fundamental defect of the model.
Ernie
With regard to which factors have failed it is interesting to note that value type factors have turned perverse in a dramatic fashion in August without notable sector biases. The level is such that it is unlikely to be a random event.
This could indicate selling by a few large Hedge fund or Quant shops, since there is a tendency for many to use some of the same factors yet the behavior would imply that they are doing the opposite of what one might expect from their following their models on the long and short side. Could it be that they are forced to sell their equity positions and buy to cover in order to avoid doing so on their less liquid credit or fixed income side of their portfolio. This has pressure on certain types of securities in an unexpected fashion.
Saw some interesting data from Lehman on this theory. Any thoughts?
Dear anonymous,
Indeed, if many models share similar factors, then in times of market stress, all of them will suffer drawdowns simultaneously. Given a drawdown, risk management technique, such as the Kelly's formula that I have discussed in this blog, would dictate an immediate reduction of portfolio size in order to preserve capital ("deleveraging" in popular parlance. But actually, if you follow Kelly's formula, even if we keep the same leverage ratio, the portfolio size will have to be reduced due to the reduction of equity as a result of loss.) Unfortunately, a simultaneous reduction of portfolio in multiple funds amounts to a fire sale, further driving down asset prices. A vicious cycle thus ensues. This was what happened when LTCM collapsed, and this may be happening again now.
1. A multi-billion-dollar fund cannot respond to high-frequency factors without incurring unacceptable trading costs in the form of market impact. Hence many if not most such factors are of a fundamental nature. Based on this consideration, I would argue that this factor model must have a longer holding period than a few days.
Wouldn't RenTech be a rather prominent counter-example? They are responsible for up to 10% of NASDAQ trading on some days.
Furthermore, you are neglecting the possibility whereby market impact is the strategy or the model.
Dear anonymous,
Just because 10% of Nasdaq trading is due to Ren Tec does not mean they are engaging in high frequency trading. Liquidating a vast number of huge positions that have been held for 1 month over the course of a day may generate the same volume.
I fail to see how market impact generated by your own trading can be profitable. Perhaps you can construct an example scenario to illustrate this strategy?
Ernie
Of course, RenTech are very secretive about how they do what they do. The fact that they are very quantitative model-driven is a well established fact (based on their hiring practices and info disclosed by Jim Simmons himself).
Furthermore, based on references in various interviews with Simmons, high-frequency approach seems to be the consensus of many observers of RenTech. Because probability distributions in the markets are constantly changing, as soon as you've found a stable pattern you need to trade on it. Conversely, a pattern needs to be only stable enough for them to act on it. RenTech can certainly do that, having direct market access at several key exchanges. It all fits.
I fail to see how market impact generated by your own trading can be profitable. Perhaps you can construct an example scenario to illustrate this strategy?
Regarding market impact being part of the strategy, the idea is actually quite old. People can't help being somewhat irrational in their reactions to news and price changes -- the fact that can be exploited as yet another market inefficiency. Recall Jesse Livermore "testing" market depth with probing orders. Quoting Jim Simmons again, "We will do experiments of various sorts ... and in some cases we actually have to move large amounts of money through the system and see what happens. In a given experiment we might pump billions of dollars through the market..."
Now I've given you enough hints to understand how market impact can be a part of a profitable strategy...
Dear Anonymous,
Yes, we are well-aware that RenTec is quantitatively driven. I have written elsewhere on this blog about them and some of their possible strategies. In fact, in that article I speculated that they employ high frequency strategies for both stock and futures trading. This is not to say, however, that the loss in their Medallion fund is necessarily due to high frequency strategies, since it is a multi-strategy equity fund. In fact, based on everything I have heard and read so far, this is a highly unlikely scenario.
Regarding using market impact to manipulate prices and create panic in order to profit from that, I believe (based on faint memory of my Series 7 exam) this is a violation of SEC rules. Perhaps readers with sounder legal knowledge can confirm or refute this?
Ernie
Dear Anonymous,
Actually, I found something on the books regarding the illegality of your market impact strategy:
Securities Exchange Act of 1934
Section 9 -- Manipulation of Security Prices
a. Transactions relating to purchase or sale of security
It shall be unlawful for any person, directly or indirectly, by the use of the mails or any means or instrumentality of interstate commerce, or of any facility of any national securities exchange, ...
1. ...
2. To effect, alone or with one or more other persons, a series of transactions in any security registered on a national securities exchange ..., ...
raising or depressing the price of such security, for the purpose of inducing the purchase or sale of such security by others.
Ernie
Post a Comment