This year marks the 100th anniversary of the publication of Frank Knight’s Risk, Uncertainty and Profit. Knight was one of the economists who built Chicago’s reputation as a powerhouse, and is widely regarded as one of the greats. The book’s influence lives on today in the idea that some risks are quantifiable and others are not.
But while the book did draw that distinction, it did not stop there. Knight’s main conclusion, in fact, has basically been ignored for the whole hundred years since it was published.
Because the book is not simply about risk and uncertainty; it is about risk, uncertainty and profit.
Knight’s point was not merely that a distinction existed. He went on to argue that: “it is this ‘true’ uncertainty, and not risk… which forms the basis of a valid theory of profit.” In other words, that the return premium associated with measurable risk is negligible. That’s because quantifiable risk can be engineered. It can be insured against. It can be diversified.
For Knight, “The only ‘risk’ which leads to profit is a unique uncertainty resulting from an exercise of ultimate responsibility which in its very nature cannot be insured nor capitalized nor salaried. Profit arises out of the inherent, absolute unpredictability of things, out of the sheer brute fact that the results of human activity cannot be anticipated and then only in so far as even a probability calculation in regard to them is impossible and meaningless.”
It’s not difficult to see why this conclusion has been ignored: it’s inconvenient. Modeling is much easier when we quantify things, even if the inputs are just estimates. So the messy reality of real life is replaced by a simplification that allows us to make decisions. Models don’t need to be perfect to be useful.
But in this simplification, something important got lost. The difference between the model – the simplification – and reality got forgotten.
Risk, uncertainty and ESG
Investment markets are much better characterized as unquantifiable uncertainty than as well-behaved risk – see the bonus material below for a longer dissection of that statement – and that’s especially true when it comes to ESG investing.
As I’ve noted previously, sustainability is rife with unquantifiable uncertainty. Objectives are nebulous. Data is sparse, and what is available is difficult to interpret and frequently not comparable across companies. The application of fiduciary duty is contentious. Rating providers differ widely in their assessments of companies.
But, as Knight pointed out, this is where real decisions get made; this is where value can be added. The messiness is a reason to focus on ESG investing, not a reason to avoid it. ESG is where the action is because of – not despite – the fact that investors are dealing with incomplete data, with unknown and shifting relationships, with changing (and widely varying) client expectations, with difficult-to-interpret science. Investment is about embracing those challenges, not about waiting for someone else to solve them.
Bonus material: the long version of the risk hierarchy…
Frank Knight differentiated between risk where the odds are known (which he referred to simply as risk) and risk where the odds are not known (which he called uncertainty). In a recent book, John Kay and Mervyn King prefer to call the latter “radical uncertainty”. Some think of this as the “unknown unknowns”.
The distinction can, moreover, be broken down further. In a 2010 paper WARNING: Physics Envy May Be Hazardous To Your Wealth!, Andrew Lo and Mark Mueller offer a modern take on Knight’s distinction, and expand the risk hierarchy from two levels to five.
In Lo and Mueller’s structure, level 1 is complete certainty. We might think of that as the soda machine: put in a dollar – receive a can of soda. Easy.
Level 2 they call risk without uncertainty, i.e. quantifiable risk. The point here is that even though you don’t know the outcome, you do know the odds. So a fair roulette wheel would be level 2: we know the probability of any given slot in the wheel being hit. There’s a well-defined distribution of possible outcomes, and we know what that distribution is. There is risk, but it’s completely quantifiable.
This is a very convenient type of risk if your job is to build a model. Sadly, there aren’t many situations outside of card or dice games and the casino where we run into level 2 risks.
Level 3 is more subtle. Lo and Mueller call it fully reducible uncertainty. This is where you don’t know the distribution of possible outcomes, but you could work it out if you had enough information.
An example of level 3 risk is provided by a game called “pass the pigs”, which has sold tens of millions of copies over the past 40 years. For those not familiar with this game, it’s just like a dice game, except that instead of dice there are small plastic pigs. Like normal dice, each pig can fall in six different positions. But the frequency of each outcome is not equal; some positions happen much more often than others.
What is the probability that a single roll of a single pig will land on its back? If you know nothing except that there are six possible outcomes, you’d most likely make an initial estimate that the probability is one in six, or roughly 16.7%. This is the same probability that you’d place on a normal dice roll being a 6, but there’s something different about the two estimates. With the pig, you not only don’t know the outcome, you’re not really sure about the odds.
But now suppose you roll a pig and it doesn’t land on its back. That’s not a surprise. If after ten rolls it’s still not landed on its back, would you still estimate a one in six chance of the next roll ending on its back? There’s good reason to revise your estimate down now. You may even whip out your old notes on Bayesian theory and work your way to a new estimate somewhere south of 10%.
Or suppose you have – as one researcher did for a paper published in the Journal of Statistics Education – performed twelve thousand pig rolls and found 22.4% of them landed on their back. You’d probably estimate the probability of the next roll landing on its back at 22.4%.
The more rolls you see, the greater your confidence in the odds. Even after millions of rolls, you still wouldn’t know what the next outcome would be, but you would feel that you know the odds by now. The nature of the problem would have changed; there’s still risk, but this would be level 2 risk.
Thus, the more data that is obtained, the closer level 3 risk gets to level 2.
It’s tempting for investors to approach the markets in that same spirit. If only we could get enough data, we feel, then we could unlock the pattern of the markets.
But we couldn’t. The behavior of markets is not fixed. Returns are not drawn from a stable distribution that’s out there waiting to be figured out. To understand the type of risk we’re really dealing with, we need to move on to Lo and Mueller’s next level: level 4, or partially reducible uncertainty.
Imagine if you can that our small plastic pigs have feelings and can move. They can change their shape. They have moods. They go through phases. They like to herd together with one another. Sometimes they flail around wildly, sometimes they freeze in very odd positions. In this case, twelve thousand rolls would be helpful information, but not as helpful as when the odds were stable. With unstable odds, we can never get to a level 2 state of risk-without-uncertainty no matter how much past data we have.
And now – at level 4 – we have reached the true nature of the risk in markets.
There is another level – level 5 – in Lo and Mueller’s hierarchy: total ignorance. Fortunately, that’s not what we are dealing with (most of the time) in financial markets. Our knowledge of markets is not zero. We know more than nothing – and in some regards we know a lot – about the patterns of market behavior, even if complete knowledge will always elude us. It is possible to build good decision processes around what we do know, even if we lack certainty.
Lo and Mueller end their description of the various levels of risk with the proposal “that the failure of quantitative models in economics and finance is almost always attributable to a mismatch between the level of uncertainty and the methods used to model it.” So while it’s convenient to concentrate on the measurable risk, and to ignore the unmeasurable, that convenience sometimes comes at a high price.