The Wisdom of Crowds and the Anatomy of a Panic

Major asset markets are generally too large to be overly influenced or pushed around by any one participant, and so are characterized as reflecting the wisdom of the crowd. This is thought of as a positive, as individual participants value different things, have different utility functions and approach and weigh the same incoming information in different ways. The net result, through different individual buys, sells and portfolio shifts gives the ‘best’ value at any one point.

We read an interesting exchange from polymath Sam Harris on the wisdom of crowds the other day that got us thinking about how this works in reality over fairly short, but significant, periods in markets. While the wisdom of crowds notion is true over the medium term, over shorter periods, some participants do cause flows that overwhelm. The events of early 2018 are a good example. The exchange is below; Harris had spoken at an event the previous night in New York with psychologist and economist Daniel Kahneman.

Frank Villavicencio: Sam. I attended & enjoyed it but didn’t get to ask my question: your take on collective decision making. Given the many identified flaws in our individual cognitive abilities, should we consider humans as more optimized for collective, swarm-like decisioning?

Sam Harris: The crowd is only wise when individual errors are uncorrelated. When correlated—as is the case when specific biases are widely shared—there’s no safety in numbers.

Hits the nail on the head. When the errors are correlated, you don’t have a crowd, you have a mob.

With regard to markets in the shorter term and faster participants, there are hundreds of ways of entering a position. Different methods of valuation, relative valuation across asset classes, technical strategies. These all have myriad variations within them. These, broadly, are ‘uncorrelated’ in terms of approach. The ‘error’ in our view is in the increase in the use of volatility as a way to size exposures in the name of risk control. Position entries tend to be uncorrelated but position exits can be highly correlated. As Harris says – at that point there’s no safety in numbers. When many use the same risk control…you get a rush for the exit that becomes self-fulfilling. Zooming in to the early part of last year – it is clear what happened. The first chart is the S&P 500, ending at the end of January 2018. A move to all-time highs – every trend follower and breakout strategy was long – and given the low volatility, many had increased exposure. Risk balancing strategies would be at the high end of exposure as well.


The details will vary, but we can use a simple model to illustrate. The blue line marks a basic trend following signal – as we said above – with markets at all-time highs in Jan 2018, most trend signals would have been long, irrespective of the timeframe. CTAs generally use a blend – some long term, some short term and a host of them in between. We use a 3 month timeframe below for clarity. The orange line scales the position size by the underlying volatility, which is shown in yellow. Again, a few ways of doing it, but this gets the bulk of it. With low volatility you get a bigger position. With higher volatility, a smaller position. The dashed red line caps a position at an arbitrary 1.5x the base position – this stops an incredibly quiet period from making the system go too nuts. Take a look at the late January/early February period. The low volatility and positive trend put the model close to a maximum position. As volatility spikes from around 7% to 25% due to a -2.1% day on Friday Feb 2nd, followed by -4.1% on Monday Feb 5th. That jump in volatility alone takes the position down from 145% to 50%.


The exposure reducing flows in our mind clearly contributed to the ferocity of the down move. As Harris said – the crowd is only wise when individual errors are uncorrelated. When a large set of players are basing risk levels on the past level of volatility, and it changes drastically – the errors are correlated and there is no safety in numbers.

We wrote on a related point to this subject a few months ago on the inherent bullish equity bias volatility targeting injects (