I have a brother who works at an insurance company doing model actuarial science. His day-to-day job is to price catastrophe risks. I told him the forfeiture logic for @BabylonLabs_io and wanted him to give me a reference from a probability perspective—for example, roughly what order of magnitude would the chance be that a verification node gets hit by forfeiture within a year?
After hearing it, he didn’t answer right away. He asked me instead what probability distribution you’re using for this forfeiture event—Poisson or Weibull? Have you tested whether it has a heavy-tail characteristic? I was stumped. He said that in the finance and insurance pricing domain, for any risk event, the prerequisite is that you first define the type of its distribution, then calibrate the parameters using historical frequencies. But for the forfeiture events in the crypto space, you don’t have enough long time-series data. Also, whether forfeiture gets triggered often isn’t an independent random event: factors like node software bugs, Bitcoin network congestion, and timestamp delays are highly correlated; in extreme cases, they can cluster and burst.
He said that in actuarial science this kind of risk is called “under-modeled risk.” The most dangerous part isn’t that the probability is high, but that the probability itself is inaccurate. When an insurance company encounters this kind of risk, it will usually outright refuse coverage or slap on an exorbitantly priced premium, because uncertainty itself is the biggest cost. The validators on Babylon are basically running this forfeiture risk with no actuarial model—people think it’s fine not because the risk is low, but because the sample size is too small to have encountered it yet.
#baby $BABY
After hearing it, he didn’t answer right away. He asked me instead what probability distribution you’re using for this forfeiture event—Poisson or Weibull? Have you tested whether it has a heavy-tail characteristic? I was stumped. He said that in the finance and insurance pricing domain, for any risk event, the prerequisite is that you first define the type of its distribution, then calibrate the parameters using historical frequencies. But for the forfeiture events in the crypto space, you don’t have enough long time-series data. Also, whether forfeiture gets triggered often isn’t an independent random event: factors like node software bugs, Bitcoin network congestion, and timestamp delays are highly correlated; in extreme cases, they can cluster and burst.
He said that in actuarial science this kind of risk is called “under-modeled risk.” The most dangerous part isn’t that the probability is high, but that the probability itself is inaccurate. When an insurance company encounters this kind of risk, it will usually outright refuse coverage or slap on an exorbitantly priced premium, because uncertainty itself is the biggest cost. The validators on Babylon are basically running this forfeiture risk with no actuarial model—people think it’s fine not because the risk is low, but because the sample size is too small to have encountered it yet.
#baby $BABY