This morning, around 9:36AM, Simon's algorithm briefly triggered the 10% loss stop-out. That valuation was calculated separately using both our NxCore price feed and by looking at the broker's net liquidation value. We did what our procedure indicated, and we stopped his algorithm and liquidated the positions.
Upon further inspection, we concluded that his portfolio had been priced during a flash crash. The securities really did trade in the market at those prices that triggered the stop. But during the market disruption this morning, some securities had wildly volatile prices that weren't meaningful valuations - the market as a whole wasn't working appropriately. In a more rational and orderly market, Simon's algorithm wouldn't have triggered the 10% loss.
We learned more than one lesson from this. The first lesson is that our rules and processes were not flexible enough to manage a flash crash. We needed to give ourselves room to manage the event when the market is misbehaving. This is important both for the contest and trading hedge fund algorithms. Flash crashes have happened before, they will happen again, and we have to be ready for them. For that purpose we've changed the process on what to do when the 10% loss is triggered. Previously we had an immediate stop-and-liquidate procedure. Going forward, we're still going to stop the algorithm promptly, but we built in a pause where we're going to evaluate more closely before we actually do the liquidation. If today's events happen again tomorrow, the algorithm's performance wouldn't be impacted.
We then gave Simon a call and talked about what had happened with the pricing and with the liquidation. We told him that we didn't think that his algorithm had been treated fairly in the flash crash environment. At our suggestion, we all agreed to re-start his algorithm on Tuesday with the same value it had on Friday - essentially a do-over, as if Monday didn't actually happen. We all like the contest, we all like having Simon in the market, and we're going to let the algorithm play out to the end of his prize period.
A couple more thoughts about the pricing that we saw this morning. Obviously, the as-traded price is what it is, and there's no changing or arguing that fact. But one can, in retrospect, debate whether that price was a meaningful valuation or a flash crash.
We all had a front row seat on what the market is like during a dislocation. The price-finding process moved in fits and starts. According to Nanex, more than 100 stocks in the S&P 500 traded in more than a 10% range in the first 20 minutes this morning. One member of the S&P 500 traded through a 90% range of its price. It wasn't an orderly market, and for some securities it was profoundly disordered.
Simon's algo gyrated through a valuation range of more than 16%, including an 11% change in 3 minutes, while his underlying securities "found" their prices this morning. His algorithm wasn't making trades. However, the legs of his hedged positions found their prices at different times, and that drove a fluctuating portfolio valuation.
We're still sorting through trade data and pricing information to see what other lessons we can learn. We want to both protect our algorithm writers, our investors, and ourselves from future flash crashes and other market dislocations. We plan on sharing knowledge about what happened so that others can plan and write defensively. We also will write some defenses into the code where we can, so that the platform can do the hard work for us all.
As you all know, we are committed to transparency. We hope this view into the day's events is helpful and informative. Your feedback is welcome.