Thursday, August 16, 2007

It's the Models, Stupid

I just can't think that, in the world that we deal with, our efforts spent on low-latency, code optimization, high throughput, best algo containers, etc ... are worthless if the models are not good.

Doesn't matter how fast you get your trades out to the exchanges if you are always on the wrong side of the trade.


©2007 Marc Adler - All Rights Reserved

2 comments:

Anonymous said...

Hey Mark, it's Jerome. We worked in NYC at nano.com many moons ago. You remember?

email me at jerome.pineau@gmail.com if you get this ok?

Anonymous said...

There is still value in trivial models if you are fast enough. If you can get an order out to sell ahead of the herd when News is released or when 4 or 6 options exchanges just trended up/down for an issue you'll make money. But probably not enough money to justify the overhead of institutional trading.

At the macro level where the big trades are done latency/efficiency is probably irrelevant as long as you can get something near the BBO - which is why anonymous block trading seems to be taking off.

That just leaves Prime Broker / DMA services. IMHO selling your DMA products based on trivial technical metrics like latency says that you are selling a commodity product.

Evidence of achieving consistent price/time priority for a model/algo is what is really required. Customers should be wary of providers who keep their models/algos secret and specify only meaningless metrics like ack times.

I think the DMA market lacks trust in the capability/efficiency of institutional models and may be searching for a simple efficient routing service, and I see no reason why that can be done at wire speed.

Just my $0.02 worth :-)

Everybody ready for Reg NMS on Monday?

ttfn

c.