Negative Skewness Incentive Problem
When agents receive positive incentives (e.g., bonuses) without bearing downside risk, they are incentivized to maximize the probability of small gains and minimize the probability of observed losses, even if this degrades the total expected return and generates hidden negative skewness—a recipe for catastrophic blowups.
THE IDEA
Simplified assumptions of constant q and single-condition stopping time
The excerpt assumes a constant q, sets q = 1 and stops the sequence at the first loss. It sketches an incentive model but does not provide the full derivation or all the conditions needed for its expectation calculation.
We can express the stopping time condition in terms of uninterrupted success runs. Let the ordered set of consecutive success runs be {F}, {SF}, {SSF}, …, {(M − 1) consecutive S, F}, where S is success and F is failure over period, with associated corresponding probabilities.
The displayed expression is incomplete: its final factor is a placeholder, so it cannot be used to calculate an expected payoff: $$\mathbb{E}[\text{payoff}] \approx \frac{\alpha}{p_{\text{loss}}} \cdot \text{(something)}$$
The intended incentive problem is clearer than the unfinished formula: an agent paid for successful periods, with no corresponding exposure to losses, may favor frequent small gains while leaving the principal exposed to rare, severe losses. Such a strategy can increase the agent’s compensation even while reducing the total expected return m. The incomplete derivation does not establish that every such contract maximizes negative skewness.
Probabilistic Sustainability and Ergodicity
Dynamic Risk Taking: If you take the risk—any risk—repeatedly, the way to count is in exposure per lifespan, or in the way it shortens the remaining lifespan.
Ruin Properties: Ruin probabilities are in the time domain for a single agent and do not correspond to state-space (or ensemble) tail probabilities. Nor are expectations fungible between the two domains. Statements on the 'overestimation' of tail events (entailing ruin) by agents that are derived from state-space estimations are accordingly flawed. Many theories of 'rationality' of agents are based on wrong estimation operators and/or probability measures.
Taleb connects this concern about ruin to the barbell strategy.
This is a special case of the conflation between a random variable and the payoff of a time-dependent, path-dependent derivative function.
Less Technical Translation:
Never cross a river if it is on average only 4 feet deep.
When to use it
Designing executive compensation packages that include clawbacks or deferred equity to align long-term risk.
Evaluating bonus structures in financial trading desks where traders can earn large bonuses from small consistent gains while exposing the firm to tail risk.
Analyzing project management incentives where teams are rewarded for meeting milestones without considering the probability of catastrophic failure.
Assessing public policy decisions where short-term metrics (e.g., GDP growth) are optimized at the expense of systemic stability.
FROM THEORY TO EVERYDAY LIFE
In practice
Indy Mac during the subprime crisis
Taleb uses Indy Mac’s failure during the subprime crisis to illustrate the incentive problem: managers can be rewarded for apparently steady profits while the firm accumulates exposure to rare, large losses. The example directs attention to who receives the gains and who bears the losses.
The river-crossing analogy
Never cross a river if it is on average only 4 feet deep. The average depth is irrelevant if the river has deep holes that can drown you. Similarly, an agent's compensation based on average returns ignores the risk of ruin hidden in the left tail.
A MOMENT TO REFLECT
Put it to work.
How can incentive structures be redesigned to ensure that agents bear downside risk proportional to the upside they capture?
In what real-world situations have you observed positive-incentive schemes that produced hidden negative skewness and eventual failure?
Why might focusing on average outcomes be dangerously misleading when dealing with path-dependent risks like ruin?
Often confused with
CONTINUE THE CONVERSATION
Source texts
- Skin In The Game Hidden Asymmetries In Daily Life (Nassim Nicholas Taleb)