but it doesn’t work for non-public information that is different from market sentiments by only a degree. If you have private information that the Russian government has a 2% chance of using a nuclear weapon on Crimea (perhaps you know they will roll a 50-sided die and use them on a 1), but you can’t tell whether the current market prices imply a 0% to 4% probability, you have no way of using your private information without performing a full analysis of the asset.
I disagree for several reasons:
your example is extremely unrealistic. When do people interested in geopolitics ever get new information expressed as a precise limiting frequency like your die example? That sort of estimate doesn’t exist outside of prediction markets.
you are setting up a strawman when you say one needs to compare a 2% to a 4% estimate: you don’t need to know the market’s exact estimate, you merely need to know whether yours is lower. Usually, estimating a bound or inequality is a lot easier than estimating an exact value...
and specifically, by the reasoning I gave before about market efficiency, estimation is easy: when you get private information, you only need to know whether it would increase or decrease probabilities on its own. All other information is already priced in but your new information is not, by definition, and hence will shift the market price in the estimated direction, allowing you to profit.
To make your example more realistic: you learn from an informant that tactical nuclear bombs came up at the latest private discussion of the Russian cabinet; you know that the market prices on Ukrainian/Russian assets implicitly assign some probability to the use of nukes, and hence some smaller probability that the Russian cabinet is discussing their use, but the market does not know that the cabinet actually has discussed the use of nukes; you do. Now, you may have no idea whether the market assigns 0.5 or 50% to the use of nukes, but its assignment is being done in the absence of this information about their discussion. All you have to do is decide: is the Russian cabinet discussing the use of nukes evidence for or against the future use of nukes, in a purely-evidential odds or decibel Bayesian sense, independent of priors or posteriors? If it’s ‘for’, then whatever the market probability is (you may have no idea what it is and no ability to figure it out), it will shift upwards; and since the prices reflect the probability, you have an opportunity to short.
If there were a prediction market it would be straightforward to do so, however.
You can do the same thing with prediction markets, assuming they’re big enough that you can treat them as efficient. Did you learn new information which is positive? Buy. Negative? Short. Knowing your own subjective probability is useful mostly when you suspect markets are inefficient and you can make a profit without learning any new information. (Typically, whenever I’m trading on a prediction market, I don’t even try to elicit my own subjective probability, I just anchor on the market probability and look for signs of bias or ignorance.)
It’s really just a question of efficiency. A market for a single asset will be less efficient than the market for that asset and 10 related questions because the information costs are lower for people who have private information that bears on those questions.
That’s not what the question was. The question was whether you were right that “to express your view, you have to become an absolute expert in any of the assets in question so that you can estimate the implied probability of current market prices”. Certainly, more targeted assets will make it easier to make money off targeted insights. But there’s no possible/impossible distinction where you can make money off your nuke insights on a prediction market but no one can make money off the same nuke insight on equity markets.
It seems we agree in many areas (you seem to disagree a great deal with my tone and examples, however), so I will focus on what appears to be the core of the disagreement. You are using a framework that assumes strongly efficient markets with respect to private information, and where most private information is of the sort that has a clear impact with respect to the priors implied on the market. I am using a framework of limited market efficiency, where only information that can be profitably exploited, because it e.g. provides a high enough Sharpe ratio, will be reflected in market prices, and where private information can often have an ambiguous relationship to the current odds implied by market prices. Note that my example was based on information of a probabilistic nature, analogous to using a novel statistical model or the like, whereas your example was based on discrete information. Note also that a novel statistical model can still have elements of discrete private information, as when hedge fund analysts set up cameras to monitor the comings and goings of hotel patrons, but where such information still cashes out in terms of a probability.
As part of my framework, the question of whether you can profitably reveal information to the market and the efficiency of said market are intrinsically linked, and this is not a form of dodging the core issue.
I agree that there is not a bright line separating possible and impossible when it comes to whether information can be profitably raised in the market, but there are clearly things that fall on one side or the other of the fuzzy line. My contention is that there are many matters of importance that one can have views and information on that fall on the “not profitable” side of the line. I will retract that you necessarily have to become an absolute expert on the asset. Technically, you only need enough expertise to correctly estimate the impact of your information, but realistically you will in most cases need to become an expert on the asset (and likely multiple assets, since you will want to long some and short some in order to extract as much value as possible) in order to create those estimates (As an aside, if you didn’t have to be an expert, there would be books and seminars about how to build investment strategies around more accurately predicting world events, because there are books and seminars on every conceivable investment strategy that doesn’t require one to be an expert. And yet you yourself presumably profitably participate in prediction markets instead of just using those same predictions to invest in capital markets.).
Is it important who becomes the next president of the United States? Many would say that it is, and it is a perennial favorite of prediction markets. Could you build a profitable investment strategy if you ONLY knew who was going to become the president a day ahead of time? You better sit down now and do a lot more work (i.e. invest in a tremendous amount of information cost, i.e. become an expert in the relevant assets), because the impact that will have on the markets is by no means unambiguous.
As a point which hasn’t been directly addressed, with respect to Lumifer’s original statement:
If you have a prediction (a “view”) on something important, you can often express that view in financial markets.
I assume that this can be translated to something like, “If you have views on something important, you can often express that view in the financial markets to achieve a higher risk-adjusted return than you could in absence of those views.” A possible problem with estimating probabilities of events which only make up a tiny portion of the expected value of a given asset, is that the expected return is totally swamped out by the volatility. You alluded to this earlier in the parenthetical:
(Your real problem is whether you can buy enough to make it worthwhile and lack of diversification & volatility means you may be right, buy appropriately, and lose anyway, but that’s why you work for a hedge fund.)
Suppose that you decide a Democratic president will win with higher probability than is expected by the market, due to either your information model or mine, and after analyzing the markets you determine that the best way to take advantage of this is to go short developed country stocks and bonds, and go long the stocks and bonds of the United States. It could easily be the case that despite having good information and the optimal strategy, the residual volatility between your hedge positions makes this a bad strategy on a risk-reward basis, because it overrides the return differential, particularly after costs. This makes it possible but unprofitable to reflect your views in the financial markets, and is a fairly fundamental issue with Lumifer’s idea, since in general we expect our views to be only marginally more accurate than the market, but we observe fairly large volatility in the differences of even correlated assets. For example, correlations between US and European stock markets are on the order of .6, which leaves a substantial amount of residual volatility if you are hedging them. Fundamentally, a perfect prediction market would require two assets which are perfectly anti-correlated EXCEPT for the scenario in question, and the further you get from that ideal, the harder it is to create profitable predictions on a risk-reward basis. And the more assets that are required to build your trade, the more assets you require expertise in, and the more you pay in fees. I would contend that this is generally the case for specific predictions that do not bear directly on the movement of major assets.
You are using a framework that assumes strongly efficient markets with respect to private information, and where most private information is of the sort that has a clear impact with respect to the priors implied on the market. I am using a framework of limited market efficiency, where only information that can be profitably exploited, because it e.g. provides a high enough Sharpe ratio, will be reflected in market prices, and where private information can often have an ambiguous relationship to the current odds implied by market prices. Note that my example was based on information of a probabilistic nature, analogous to using a novel statistical model or the like, whereas your example was based on discrete information.
OK. And I think you are wrong in thinking that markets are limited in efficiency and that clearly relevant private information nevertheless has ambiguous implications.
I agree that there is not a bright line separating possible and impossible when it comes to whether information can be profitably raised in the market, but there are clearly things that fall on one side or the other of the fuzzy line.
Transaction costs, available capital, risk, market efficiency, and other factors set a fuzzy line for what kind of information and how large a change in probability will be profitable, yes. We’re dealing with real world markets and events, after all.
Technically, you only need enough expertise to correctly estimate the impact of your information, but realistically you will in most cases need to become an expert on the asset (and likely multiple assets, since you will want to long some and short some in order to extract as much value as possible) in order to create those estimates
I think most news is fairly easily evaluated. The Russian atomic bomb example may be too clean an example, but it doesn’t seem terribly hard to guess whether Steve Jobs unexpectedly going to a hospital is bad or good for APL.
Is it important who becomes the next president of the United States? Many would say that it is, and it is a perennial favorite of prediction markets. Could you build a profitable investment strategy if you ONLY knew who was going to become the president a day ahead of time? You better sit down now and do a lot more work (i.e. invest in a tremendous amount of information cost, i.e. become an expert in the relevant assets), because the impact that will have on the markets is by no means unambiguous.
I’ve already linked to papers which have done the footwork for one interested in that question. It’d take maybe an hour to read them. Is that ‘a lot more work’? And how hard would a finance professional with access to the relevant databases find to replicate the analysis, even if they had no a priori beliefs about how a Democratic victory might affect various stocks?
It could easily be the case that despite having good information and the optimal strategy, the residual volatility between your hedge positions makes this a bad strategy on a risk-reward basis, because it overrides the return differential, particularly after costs. This makes it possible but unprofitable to reflect your views in the financial markets, and is a fairly fundamental issue with Lumifer’s idea, since in general we expect our views to be only marginally more accurate than the market, but we observe fairly large volatility in the differences of even correlated assets. For example, correlations between US and European stock markets are on the order of .6, which leaves a substantial amount of residual volatility if you are hedging them.
I don’t follow this point. Why can’t I find an appropriate set of hedges? Doesn’t that imply inefficiencies?
Fundamentally, a perfect prediction market would require two assets which are perfectly anti-correlated EXCEPT for the scenario in question, and the further you get from that ideal, the harder it is to create profitable predictions on a risk-reward basis.
I assume you mean by ‘perfect’ a prediction market in which even the slightest bit of new evidence can be profitably exploited because there are no kinds of transaction costs or other friction? That may be true, but I don’t think it meaningful refutes Lumifer’s observation that markets allow for expression of views on “something important”.
I disagree for several reasons:
your example is extremely unrealistic. When do people interested in geopolitics ever get new information expressed as a precise limiting frequency like your die example? That sort of estimate doesn’t exist outside of prediction markets.
you are setting up a strawman when you say one needs to compare a 2% to a 4% estimate: you don’t need to know the market’s exact estimate, you merely need to know whether yours is lower. Usually, estimating a bound or inequality is a lot easier than estimating an exact value...
and specifically, by the reasoning I gave before about market efficiency, estimation is easy: when you get private information, you only need to know whether it would increase or decrease probabilities on its own. All other information is already priced in but your new information is not, by definition, and hence will shift the market price in the estimated direction, allowing you to profit.
To make your example more realistic: you learn from an informant that tactical nuclear bombs came up at the latest private discussion of the Russian cabinet; you know that the market prices on Ukrainian/Russian assets implicitly assign some probability to the use of nukes, and hence some smaller probability that the Russian cabinet is discussing their use, but the market does not know that the cabinet actually has discussed the use of nukes; you do. Now, you may have no idea whether the market assigns 0.5 or 50% to the use of nukes, but its assignment is being done in the absence of this information about their discussion. All you have to do is decide: is the Russian cabinet discussing the use of nukes evidence for or against the future use of nukes, in a purely-evidential odds or decibel Bayesian sense, independent of priors or posteriors? If it’s ‘for’, then whatever the market probability is (you may have no idea what it is and no ability to figure it out), it will shift upwards; and since the prices reflect the probability, you have an opportunity to short.
You can do the same thing with prediction markets, assuming they’re big enough that you can treat them as efficient. Did you learn new information which is positive? Buy. Negative? Short. Knowing your own subjective probability is useful mostly when you suspect markets are inefficient and you can make a profit without learning any new information. (Typically, whenever I’m trading on a prediction market, I don’t even try to elicit my own subjective probability, I just anchor on the market probability and look for signs of bias or ignorance.)
That’s not what the question was. The question was whether you were right that “to express your view, you have to become an absolute expert in any of the assets in question so that you can estimate the implied probability of current market prices”. Certainly, more targeted assets will make it easier to make money off targeted insights. But there’s no possible/impossible distinction where you can make money off your nuke insights on a prediction market but no one can make money off the same nuke insight on equity markets.
It seems we agree in many areas (you seem to disagree a great deal with my tone and examples, however), so I will focus on what appears to be the core of the disagreement. You are using a framework that assumes strongly efficient markets with respect to private information, and where most private information is of the sort that has a clear impact with respect to the priors implied on the market. I am using a framework of limited market efficiency, where only information that can be profitably exploited, because it e.g. provides a high enough Sharpe ratio, will be reflected in market prices, and where private information can often have an ambiguous relationship to the current odds implied by market prices. Note that my example was based on information of a probabilistic nature, analogous to using a novel statistical model or the like, whereas your example was based on discrete information. Note also that a novel statistical model can still have elements of discrete private information, as when hedge fund analysts set up cameras to monitor the comings and goings of hotel patrons, but where such information still cashes out in terms of a probability.
As part of my framework, the question of whether you can profitably reveal information to the market and the efficiency of said market are intrinsically linked, and this is not a form of dodging the core issue.
I agree that there is not a bright line separating possible and impossible when it comes to whether information can be profitably raised in the market, but there are clearly things that fall on one side or the other of the fuzzy line. My contention is that there are many matters of importance that one can have views and information on that fall on the “not profitable” side of the line. I will retract that you necessarily have to become an absolute expert on the asset. Technically, you only need enough expertise to correctly estimate the impact of your information, but realistically you will in most cases need to become an expert on the asset (and likely multiple assets, since you will want to long some and short some in order to extract as much value as possible) in order to create those estimates (As an aside, if you didn’t have to be an expert, there would be books and seminars about how to build investment strategies around more accurately predicting world events, because there are books and seminars on every conceivable investment strategy that doesn’t require one to be an expert. And yet you yourself presumably profitably participate in prediction markets instead of just using those same predictions to invest in capital markets.).
Is it important who becomes the next president of the United States? Many would say that it is, and it is a perennial favorite of prediction markets. Could you build a profitable investment strategy if you ONLY knew who was going to become the president a day ahead of time? You better sit down now and do a lot more work (i.e. invest in a tremendous amount of information cost, i.e. become an expert in the relevant assets), because the impact that will have on the markets is by no means unambiguous.
As a point which hasn’t been directly addressed, with respect to Lumifer’s original statement:
I assume that this can be translated to something like, “If you have views on something important, you can often express that view in the financial markets to achieve a higher risk-adjusted return than you could in absence of those views.” A possible problem with estimating probabilities of events which only make up a tiny portion of the expected value of a given asset, is that the expected return is totally swamped out by the volatility. You alluded to this earlier in the parenthetical:
Suppose that you decide a Democratic president will win with higher probability than is expected by the market, due to either your information model or mine, and after analyzing the markets you determine that the best way to take advantage of this is to go short developed country stocks and bonds, and go long the stocks and bonds of the United States. It could easily be the case that despite having good information and the optimal strategy, the residual volatility between your hedge positions makes this a bad strategy on a risk-reward basis, because it overrides the return differential, particularly after costs. This makes it possible but unprofitable to reflect your views in the financial markets, and is a fairly fundamental issue with Lumifer’s idea, since in general we expect our views to be only marginally more accurate than the market, but we observe fairly large volatility in the differences of even correlated assets. For example, correlations between US and European stock markets are on the order of .6, which leaves a substantial amount of residual volatility if you are hedging them. Fundamentally, a perfect prediction market would require two assets which are perfectly anti-correlated EXCEPT for the scenario in question, and the further you get from that ideal, the harder it is to create profitable predictions on a risk-reward basis. And the more assets that are required to build your trade, the more assets you require expertise in, and the more you pay in fees. I would contend that this is generally the case for specific predictions that do not bear directly on the movement of major assets.
OK. And I think you are wrong in thinking that markets are limited in efficiency and that clearly relevant private information nevertheless has ambiguous implications.
Transaction costs, available capital, risk, market efficiency, and other factors set a fuzzy line for what kind of information and how large a change in probability will be profitable, yes. We’re dealing with real world markets and events, after all.
I think most news is fairly easily evaluated. The Russian atomic bomb example may be too clean an example, but it doesn’t seem terribly hard to guess whether Steve Jobs unexpectedly going to a hospital is bad or good for APL.
I’ve already linked to papers which have done the footwork for one interested in that question. It’d take maybe an hour to read them. Is that ‘a lot more work’? And how hard would a finance professional with access to the relevant databases find to replicate the analysis, even if they had no a priori beliefs about how a Democratic victory might affect various stocks?
I don’t follow this point. Why can’t I find an appropriate set of hedges? Doesn’t that imply inefficiencies?
I assume you mean by ‘perfect’ a prediction market in which even the slightest bit of new evidence can be profitably exploited because there are no kinds of transaction costs or other friction? That may be true, but I don’t think it meaningful refutes Lumifer’s observation that markets allow for expression of views on “something important”.