AlgoTrading 2005 Conference
Earlier this week I had an opportunity to sit in on a bunch of sessions at
the AlgoTrading2005 conference, and overall it was worth the trip. Some
interesting "take-aways" including:
- Seemed to be a consensus that there was an 80/20 rule at work with algo
trades. 20% of the orders a b/s firm had to do are “hard” and need to be worked
and consume 80% of time/resources whereas 80% of the trades are “easy” and could
be done using algo, crossing or some other low touch solution. Most folks were
saying it’s 50/50 now but would grow to 80/20. - Still a big need for post-trade analytics and TCA. Post-trade need is more
of a tactical “how did I do on this trade [relative to how I expected to do]”
and TCA is more strategic “how did I do this quarter in terms of execution
cost?” - There are currently 28 s/s firms that offer algos. My read was that the
average s/s firm has 3 or 4 models, so some quick math and I figure that there’s
something like 100+ models available on the street. I asked a b/s guy how he
evaluates models and how many he uses at any given time and he said, “2 or 3.”
It's like the auto industry before Henry Ford showed up…mass customization. - From the b/s, what are the key differentiators? (1) Easy to understand, (2)
fits every trading style, (3) post-trade analytics, (4) real-time alerts, (5)
speed, (6) customization. The last point was repeated very frequently. - Of the above mentioned points, probably #1 (ease of use) and #6
(customization) were the most repeated. b/s traders talked a lot about things
like “comfort with a model” “simplicity” “familiarity” and things like that.
With respect to “customization” they talked about being able to “tune” models.
Having settings like “passive, moderate or aggressive” is already included in
some models. They wanted more of this. - The s/s is investing a lot of money in pre-trade and real-time analytics
(things GSCO’s pre-trade graphs and SmartAlerts). The problem is that the s/s
doesn’t have the “context” of the trade so they don’t know when to say stop or
go (a good point made by Will Sterling from UBS). They’re looking for ways to
push out this info to the b/s, but OMS vendors suck at delivering this kind of
info. As a result there is a shift to using “EMS” or Execution Management
Systems like FlexTrade or EdgeTrade. I think there is an opportunity to use IM
which leverages the s/s’s investment in analytics and projects that out to the
b/s using IM. - Many b/s firms (and s/s) talked about the necessity to have a checklist to
go through when evaluating algos. I'd be curious to see such a list. - The b/s said they didn’t want, “3, 4, 5 or more” front ends for algo…they
want a “one stop shop.” That's obvious. - Andy Silverman (GSCO) talked about a new model called “Dynamic Scaling”
which had a dynamically changing “aggressiveness factor” that would “buy more if
the stock moved for you and by less if the stock moved against.” Talked about
using benchmarks other than price (e.g. volume) too. I feel a little bit like
the models that the s/s is putting out are really basic. How about using some
AI? - There was some question about whether algo was a “commission” sale or a
“software” sale. I think it's a "commission" sale. It's a service sale and the
software comes and goes with each new generation of improvement. - No one asked this question, but I’m curious what the “shelf life” of a model
is. Some s/s firms talked about “4th generation VWAP” and algo has only been
around for 3 or 4 years, so we’re talking 1 year shelf life max? - The “Goldman Cube” is a way to evaluate trades on three axes: urgency, size
and liquidity. For example, the hardest trades are “very urgent, large in size
in an illiquid name” and conversely a “not urgent, small block of a liquid name”
is easy. Goldman “describes” their algos based on the Cube.
We need something like this on an industry basis…MorningStar, where are you?!?
- A guy from Merrill made an interesting comment. He said that “90% of our US
clients have FIX, but only 30% of our European clients are on FIX.” Not sure if
that’s peculiar to his group at ML, ML in general, or the industry as a whole.