Why companies pass on good bets: loss aversion among managers

A project needs $100 million. It might return $400 million in present value within three years, and it might lose all the capital in the first year. What is the highest chance of loss at which you would still go ahead? A risk-neutral person says 75%, because a quarter of $400 million is exactly the $100 million put in. In 2012 this question was put to 1,500 managers. Most would accept only 18% (Lovallo et al., 2020).

How risk-averse are managers, according to that survey?

According to Lovallo, Koller, Uhlaner and Kahneman (2020), Koller and Lovallo put the scenario to 1,500 managers in a 2012 McKinsey global survey. Only 9% would accept a chance of loss of 40% or more. When the capital was cut to $10 million with a possible $40 million return, they were just as cautious, so the size of the investment made little difference (Lovallo et al., 2020).

My own arithmetic shows the gap. Accepting an 18% chance of loss means going ahead only when the chance of success is 82% or more. The expected value is then 0.82 times 400 less 100, which is $228 million, more than twice the capital. Taken as a group threshold, 18% behaves as if they wanted an expected gain that large before saying yes, where the risk-neutral level is an expected gain of zero.

According to the summary of the article, corporate incentives and control processes actively discourage managers from taking risks (Lovallo et al., 2020).

What is loss aversion?

It is the observation that a loss weighs more than an equivalent gain. Kahneman and Tversky (1979) write that the aggravation of losing a sum of money appears greater than the pleasure of gaining the same amount, and that most people find symmetric bets, win x or lose x with equal probability, distinctly unattractive (Kahneman and Tversky, 1979, p. 279). In their model the value function measures changes from a reference point, and is steeper for losses than for gains (Kahneman and Tversky, 1979, p. 275, 279). The idea behind the term loss aversion in the site library comes from here.

The reference point is usually the current position, but the paper also says gains and losses can be coded against an expectation or aspiration level that differs from the status quo (Kahneman and Tversky, 1979, p. 286). For managers the closest example is the annual target. My inference: a project that leaves the budget short of target is booked as a "loss", whether or not the spending was worth it.

Do not read loss aversion as risk aversion everywhere. The same paper records that between losses people tend to seek risk, preferring a gamble that might lose more to a smaller sure loss (Kahneman and Tversky, 1979, p. 268). My inference: a manager who is behind target may keep pouring money into an old project that is losing rather than book the loss.

Why do companies amplify loss aversion?

Because the person who approves a project and the person who bears its risk do not see the same picture. According to the public summary of Lovallo and colleagues, theory says companies create value through risky investment, but in practice managers in large firms routinely quash risky ideas in favour of marginal improvements, cost-cutting and "safe" spending. A CEO considers each investment within a wider portfolio, whereas managers bet their careers on each one, even when the outcome is negligible for the corporation as a whole (Lovallo et al., 2020).

The article opens with the condition behind the argument. Investments can be large, and even a significant percentage of failures does not matter, so long as no single failure sinks the enterprise and the successful bets compensate, which usually happens. It rests on Markowitz's portfolio theory from the 1950s (Lovallo et al., 2020).

The authors call the gap between the choices a CEO would make and the choices managers actually put forward a "risk aversion tax" (RAT). At one high-performing company they assessed all investments made in a year and calculated a RAT of 32%: the firm could have improved its performance by nearly a third just by deciding according to the CEO's risk tolerance rather than that of junior managers (Lovallo et al., 2020).

How does a portfolio of ten projects look different from one project?

A wholly hypothetical example, with numbers of my own, meant to connect the survey to an SME. Each project needs VND 500 million. If it succeeds it returns VND 2 billion, capital included (the same 4-to-1 ratio as the survey). If it fails, the VND 500 million is lost. The chance of success is 40%.

A single project fails 60% of the time. That is below the risk-neutral threshold of 75%, so a risk-neutral person goes ahead, and a person who accepts only 18% declines. Across ten projects the company loses overall when two or fewer succeed, a probability of about 17% (my own binomial calculation), so there is an 83% chance of a profit. The manager proposing one project sees a 60% chance of her name being attached to a failure. The owner looking at ten sees an 83% chance that the whole portfolio makes money. The gap between those two views is where the risk aversion tax arises.

Two assumptions are open to challenge. The ten projects must be independent, whereas projects exposed to the same market shock make an overall loss far more likely. And the company must be large enough to survive a run of failures.

What can be done, and when is risk aversion reasonable?

The article proposes measuring first. In the excerpts I read, a company can estimate the tax by surveying the risk tolerance of the CEO and of managers at several levels (Lovallo et al., 2020). The authors write that companies need to switch from processes built around managing outcomes to ones that encourage a rational calculation of the probabilities (Lovallo et al., 2020). In your own firm, try this: choose a sample project, ask each manager the highest chance of losing the capital that they would still accept, and compare it with 1 minus capital divided by return. The threshold in your firm is yours to set. To choose projects in the first place, see NPV or IRR. To record how you estimated, see the decision journal.

Risk aversion is reasonable when a single failure is enough to sink the business. The argument of Lovallo and colleagues stands only when that condition does not hold (Lovallo et al., 2020). The owner of a small workshop seldom has ten projects, and VND 500 million could equal a full year's profit. My inference: for a small business, the question should be what is left of the company if this project loses, not whether managers are being risk-averse.

Going deeper: what I read and what is missing

I read Kahneman and Tversky (1979) in full, from a university's archived copy, with page numbers following the journal. The experiments rest on hypothetical questions put to students and university faculty (Kahneman and Tversky, 1979, p. 264). The authors themselves note that reliance on hypothetical choices raises questions about validity and generalisability (Kahneman and Tversky, 1979, p. 265).

Lovallo and colleagues (2020) sits behind a paywall. I read the summary, the opening paragraph and long excerpts of the article reproduced by an economics blog. I have not read the whole article, so I do not know its full recommendations. I also do not know who the survey respondents were, or whether 18% is a mean or a median. Both big figures, 18% and 32%, are the authors' own: the survey was a hypothetical question, and the RAT is an example from one company. The RAT calculation also assumes that the CEO's risk appetite is the right one. That is an assumption, not a measured result.