Most of the time when an organisation insists on you signing multiple NDAs before you are even allowed past the front door, it's a con. Whatever it is they are showing you (in AT's case usually a trading model concept or a piece of technology) is probably junk and therefore needs the maximum amount of window dressing to appear credible.
In this case, the omens weren't particularly promising. The premises were in a seriously run down part of a rather unexpected city. The office foyer looked (and smelt) like it had a part time role as a rest home for the local feral cat population, while the doorman had clearly graduated summa cum laude joint honours in surliness and halitosis.
And yet there was something in the manner of the young man
thrusting a sheaf of NDAs at us that was rather more interesting.
He wasn't somehow slick enough to be the usual smoke and mirrors
merchant. Either that or he was good enough to know not to be.
Having negotiated his multiplicitous legal documentation, we
proceeded to the basement. Things started to look up. The
fixtures and fittings budget went from negative to stately home
in about three paces - and the technology capex considerably
further. Somebody had spent some very serious cash on hardware,
but very little on human assets - of which there appeared to be
Not so, as various individuals gradually emerged from among the server racks and a meeting of sorts began. Our first attempts to pigeonhole this curious organisation were well wide of the mark. Hedge fund? Prop shop? No. Nearest we got to a consensus description was a privately funded capital markets research institute.
So what were they researching? It was AI of a sort, but taken to the nth degree. Every element in the trading process was effectively self building. The design team weren't constructing trading models, but building the technology that would completely automate their construction. The only initial human input to the process was the desired outcome in terms of rates of return, variance, stability etc. The obvious proviso was that the more optimistic the inputs, the longer the wait.
But it went beyond that to include auto building and refining back office technology as well. As models and assets were added they could be aggregated and the technology would self build the optimal back office infrastructure for that particular combination. It would then adapt the infrastructure in real time in accordance with how the models were trading. For example, if they started to include more fixed income in their trading, then the back office infrastructure would be automatically tweaked to allocate more computing resources to processing fixed income trades.
We were shown some real time performance figures. Assuming they weren't as surreal as the organisation, there seemed some serious merit to the group's activities. One of the most intriguing aspects of the self build process was the approach it took to trading model life span, which was negligible. Trading models were being created in their thousands, traded for perhaps just one trade and then discarded. As we were wafted back on to the street on the wings of the doorman's fragrant breath, we wondered whether we had witnessed the future or just an elaborate scam - and also why we had been invited at all. No publicity motive certainly, the NDAs ensured that, but perhaps just a desire to impress an audience?