Quite some time ago, in 1921, Frank Knight, an American economist from the University of Chicago, wrote, in his highly influential book, “Risk, Uncertainty and Profit”, that we should worry about uncertainty, not risk. His assertion was grounded in the definition of each term. He defined risk as randomness that can be measured by a known probability distribution, and contrasted it with uncertainty, that is the randomness that cannot be measured. Up to now the paradigm states that since risk can be measured, it could be hedged, and we should not worry about it. Consistent with this, Knight posits that the existence of uncertainty (not risk) is the reason why firms generate a profit that cannot be offset by perfect competition. This whole discussion is, at least to me, quite provoking, and, interesting enough to get me started on this post.
My main problem with the abovementioned paradigm is the assumption that we can always measure risk, I tend to question it. Can we really do that? What is our purpose? In his book, Knight writes: “The aim of science is to predict the future for the purpose of making our conduct intelligent”. If the reason why we want to measure risks is the one proposed by Knight, then we have to question the rationality of using past distributions to predict the potential future outcomes, at least for some variables. We might be able to measure the past evolution of oil prices, and obtain a very good fitted distribution, but we need to recognize the possibility that the future evolution of prices follow a completely different path than the one we predicted observing past data, as it actually happened and everybody learned in the past few years…
People tend to have a bias towards precision; we prefer to have a precise measurement rather than an imperfect estimation of anything. Unfortunately, when we measure risks from past data, something that we usually do using mathematical time series models, we tend to rely on the model fitting the data. In other words, when we obtain a fit with a given level of statistical certainty, we feel comfortable and believe that if we know the past distribution of the variable, we have successfully measured risk, and thus we can hedge it. At this point is when we usually get into trouble, because we tend to think that the risk is under control. The correct statement is: risk would be under control, if the future behavior of the variables were the same that we observed during the period of measurement, which might not be the case.
In reality, there is no way we can predict the future behavior of a random variable with certainty, unless we know the real determinants of its behavior. In our aim towards the perfect equation that explains the past behavior we tend to pass over the truly important piece of information we need, that is, which are the determinants of a given variable or Risk Factor (RF)? In other words, we need to know that RF is a function of x, y, z, p and an error. In symbols it would be,
RF: f (x, y, z, p, e)
I like to tell my students that this is the expression of our ignorance; in the sense that we give up… we surrender to the evidence that we do not know the real functional form of the function explaining the behavior of RF. Thus, we set for a second best, finding the determinants of the behavior of the variable and trying to understand the nature of their effects on RF.
If we devote enough efforts in trying to learn the determinants of the risk factors affecting our company, we can then monitor their evolution and learn about their own determinants, which would lead to a better understanding of our risks and a more accurate prediction of their future behavior. Some of the risks that we though were impossible to measure suddenly turn into measurable risks; we might have some error in our estimation but we will get the general direction right, which is extremely valuable, especially in some type of risks.
Consider the case of the price of a given commodity of your choice, let’s say soybean. We can easily agree that its price is influenced by supply and demand. The definitions of supply and demand, however, might not be easy to agree on. Are they only physically driven? Or we also need to consider their financial component? In fact, the prices of most of the commodities that are actively traded in the financial markets are driven by their fundamentals, i.e. the physical demand and supply of their producers and customers, but they are also driven by the technical, i.e. their supply and demand by financial investors (which has completely different drivers). A short example might helps us understanding this discussion. Towards the end of 2012, soybean was trading at high prices in the Chicago Mercantile Exchange. The fundamentals were suggesting support for those high prices: (i) demand from China was steady and increasing, and (ii) supply from the American producers was not guaranteed: both, South America (mainly Argentina and Brazil) and North America, were facing severe climate conditions that casted doubts on the final figures of the crop. These strong bullish “fundamentals”, however, contrasted with the bearish “technicals”: (i) the investment funds’ holdings of soybean where the highest in a years, (ii) the soybean prices had been increasing steadily, providing hefty profits for the fund managers that might be willing to take some of these profits and reduce their exposures.
Failing to understand these determinants of soybean prices in full might fool analysts that are relying solely on past data to measure risks. There might be several variables that are determinants of soybean prices but are dormant during several periods of time (i.e. their coefficient in a regression would be equal to zero), but in a given macroeconomic situation they wake up and produce an effect in the model that has not been expected in advance. An example of this is the role played by political issues in commodity prices. A government influencing the production or the commerce of a given commodity has the potential to change prices behavior in an unforeseen way. Usually, when the event happened is easy to see how obvious it was, the problem is how do we anticipate it, instead of stating the ex-post logic…
In other words, failing to understand the complete set of drivers that determine the price of a commodity could have easily misled our judgment regarding its future price evolution. Most important, just by looking into the past data alone might not by of any help, unless we can understand the complete set of drivers to include into the econometric model.
The commodity boom of 2007/2008 can be explained by what we discussed in this post. Several, previously unknown, factors; (i) an unprecedented macroeconomic turmoil that left investors with less financial options to invest in, (ii) new financial instruments that allowed more (less sophisticated) investors in the commodity markets (Exchange Traded Funds –ETFs- are a good example), (iii) the rise of several platforms to trade online that favored day (and noise) traders to enter and exit the market with lower barriers. The conjunction of these factors, unknown before, produced a jump in prices that no model had been able to predict. Unfortunately, investors and firms relying on those mathematical models found themselves with prices that nobody could forecast and faced severe losses. The problem here is not, having failed to forecast the future price behavior; the problem is failing to understand that the models were structurally unable to forecast prices, and relying on the false sensation of being hedged. Using this framework is easy to understand how the models failed to predict most of the macroeconomic variables that generated the subprime crises starting in the summer of 2007, with its subsequent impact in currencies and the global economy.
Notice that through all this post I used examples of risks that are extremely hard to forecast. We can also use examples of risks that we can forecast without major problems, mainly those variables for which we have a good understanding of the drivers behind their volatility. Notice that the main issue is always the same: understanding the determinants behind the volatility is the key to forecast accurately.
When we think we do not face a certain risk, mainly because we were able to hedge it properly, we become bolder in the assumption of other risks. The problem is that sometimes it can be true that, not only we are not hedged, but also, we are facing more risk than expected, and on top of that, we add more operational risks, based on a false sensation of security.
