Reputation and staking: trusting an agent

Graduate · enrolled learners

This lesson opens with The credit rating agencies, 2007 to 2008.

What happened

In the years before the 2008 crisis, the major credit rating agencies rated large volumes of mortgage-backed securities and related products at the highest grade, effectively certifying them as very safe. Investors worldwide trusted those ratings and bought accordingly. The ratings were catastrophically wrong: when the housing market turned, securities rated safe defaulted in waves, and the failure of those ratings was a central mechanism of the crisis. A core reason the ratings were unreliable was that the agencies had no skin in the game. They were paid by the issuers of the securities they rated, and they bore no loss when a rating proved wrong, so a good reputation for accuracy was something they could spend without paying for being inaccurate. Years later, one agency paid about 1.5 billion dollars and another several hundred million to settle claims over those ratings, but the reckoning came long after the trust had already been given and the losses taken. The agencies had reputations and no stake at risk, and a reputation with nothing behind it turned out to be worth what it cost to give: nothing.

The decision point

Trusting an agent, human, institutional, or software, to act honestly requires a reason to believe it will bear a cost if it does not, and a reputation alone does not supply that reason if the reputation carries no stake. The credit rating agencies are the case: they had strong reputations and issued ratings that others relied on, but because they bore no loss when a rating was wrong, the reputation was cheap to give and unreliable to trust, and it failed at the worst moment. This is why systems for trusting autonomous agents lean on staking, not just reputation: an agent that must post a stake it forfeits if it misbehaves has skin in the game, so defecting costs it something real, and its good behavior is backed by more than a score it can spend freely. A reputation score records what an agent has done; a stake makes misbehaving expensive, and only the second changes the agent's incentives. So the decision when trusting an agent to transact is to ask what it stands to lose if it defects, whether there is a stake at risk and not merely a reputation asserted, because the rating agencies showed that a reputation with no cost of being wrong is worth nothing exactly when it is relied on most.

Recorded loss

$1,500,000,000

Orientation and Year One are open: anyone can read them without an account. From Year Two onward the lessons are for enrolled learners, because progress through the later years only means anything if it is tracked against a record.

It is free. We do not sell the list and there is nothing to buy at the end of it.