Testing For Scalability
As it relates to the testing of trading systems, I use “scalability” to describe a concept, although that concept is something I have difficulty testing directly.
The concept of scalability is the general question of how the amount of money managed impacts the returns. More specifically, for each system (or management style, but I really think of those as “insufficiently documented systems”), there are a series of interrelated questions that must be answered. How much money is needed to execute the system with sufficiently low risk of ruin? (which begs another question … ) At what point should the manager stop adding funds to a system? What is the relationship between system returns, volatility, and assets under management, and how does the manager capitalize on this relationship?
In my limited experience, I’ve come across a range of systems with drastically different scalability, but just about every system I’ve come across has some kind of “sweet spot.” Long-term trend-following (LTTF) systems, when executed against a large basket of futures contracts, have a fairly high hurdle for retail traders to use them safely, and run out of room at points I would consider – but that many institutions would NOT consider – stratospheric. Microcap momentum, small cap value, and NCAV systems have sweet spots that only the smallest of fund managers can hit, but any schlub with a retail account can execute these strategies. Other systems may be near infinitely scalable on the large end, but require billions (instead of the mere million-dollars-plus that LTTF requires for relative safety) to execute with a low risk of ruin.
The problem I have in testing is determining exactly what that sweet spot is!
Here’s an example test – running stock trading system X with a list of qualifiers sorted by characteristic Y, with minimum market cap filtered at $1 billion.
I can simulate retail accounts by holding 5, 10, 15, or 20 stocks at a time. I can put together some risk-adjusted return curves by testing more holdings, say 25, or 30 stocks at a time, and from there I may find that, for THIS retail trader, maybe 10 to 20 at a time is the “right” number. But is the idea scalable? If the dropoff in risk-adjusted performance is drastic, then maybe not, but if it isn’t?
I can repeat the test holding 50 or 100 of the top qualifiers. Repeat the test holding ALL qualifiers (effectively not sorting candidates by characteristic Y anymore), and see what the average number of qualifiers, and the minimum number of qualifiers, was over the test period.
I can do all of those same tests with different market cap minimums, like $100 million or $500 million, and see if the numbers change.
I may determine that, for a system X where holding 20 or so stocks looks “just right” to me as “Joe Schmuckatelli, Retail Trader,” that a fund manager using that same system X could potentially hold an average of 344 stocks with market cap $1 billion or more, with 18% monthly turnover, over a test period of 11 years, with outstanding (for a fund manager, not necessarily for a retail trader) results.
While that’s a non-trivial answer, it doesn’t really get directly at the question!!! How much money can that system handle? At what point of AUM does a manager start moving the market, when playing in a sandbox that holds 344 stocks over $1 bil in cap, turning them over at 18% a month?


August 13th, 2008 at 9:06 am
I think you have to look at average daily volume and bid/ask sizes. If you are trading far below the normal bid/ask size, you will have almost no effect on prices (unless you are trading very quickly). For example, I think you could do intraday trading on /ES up to about 100-150 contracts with no noticeable effect on prices, because the normal bid/ask size is about 500-1000. For overnight systems that trade at the open, you probably want to be below 5% of the normal opening volume to have no effect on prices. But again, it depends on the bid/ask size. If you can look at market depth, you can see how much the price will move for a given size market order. If you have some way of looking at the market depth around the opening cross for a number of stocks over several days, you could work out how much you can expect to move the market for a given volume of trading (normalized to normal stock volume in some way).
When you get really large, you would have to stop trading everything at the open. The system might still work, but you would have to execute the orders throughout the day to minimize slippage. There is a lot of academic work on how to trade large volume without moving prices. I mostly haven’t read it because it doesn’t affect me yet. Although if everyone uses the same algorithm, you could try to detect when large buyers are consistently buying/selling all day. Which could be useful information.
August 13th, 2008 at 8:54 pm
I haven’t done an exhaustive analysis of intraday stock volumes, and I’m a far ways away from moving any markets, even when I’m grabbing some illiquid small-caps. It’ll be decades before I have to worry about this on my account.
It’s a given that trading systems on size demands modifying the execution pattern from “market at open” to hiring a trader or using a computer, but what I’m really pondering here is the upper limits of AUM execution with stock-only systems (of the kind I devise) being handled by fund managers at large scale.
I’ve seen reference to the academic papers on algorithms, but I’d really be interested in chatting with those who have done it, to get a handle on how much could be handled. Even ballpark estimates, like the one I’m working up for the next post on the subject.
Es are neither here nor there. 100 contracts represents, unlevered, how many dollars of stocks done intraday on a bid/ask basis???? For timeframes I consider tradable given my edge, I consider systems built on index or currency futures to be infinitely scalable, for all practical purposes. It would take an idiot the size of a Jerome to get too big for the market in those things …
August 14th, 2008 at 6:05 am
[…] an effort to resolve the question mentioned yesterday, Testing for Scalability, I pulled a random sampling of U.S. stocks to look at their volume turnover […]
August 14th, 2008 at 3:43 pm
100 ES contracts unleveraged right now represents about $6.5 million. But almost anyone daytrading index futures is going to be leveraged at least 3:1, so you could easily hit the limit with $1-2 million. I’ve tested some strategies that the math says to leverage 50:1, which would be trading 100 contracts with $130k. I’m not going to trust my math that much (especially since those tests only use 3 weeks of data), but there are some hedge fund managers out there that do. It does give you some impressive returns until you blow up.
The answer is different if you are not daytrading. But if you want your orders to execute at open, your size limits are going to be similar to the intraday limits. The order book might have a little more liquidity at open than it has in general throughout the day, but probably not much more.
As far as how much AUM you could handle, the average dollar volume on the NYSE for the past month looks like it is about $40B. An 18% monthly turnover is about 1% daily, so you might be able to get to about $4B if you can trade 0.1% of the total market volume. NASDAQ has more volume than NYSE, so the limit is probably higher.
You can look at the AUM for Vanguard funds and see what the annual turnover is. The PRIMECAP fund has $32B with an 11% turnover and a median market cap of $37B and it is closed to new investors.The Capital Opportunity fund is also closed to new investors with about $9B, a 14% turnover rate, and a median market cap of $16.3B. Which suggests that the amount of trading you can do without affecting returns is much lower than 0.1%, at least according to the managers at Vanguard. It looks like they don’t
want to trade more than 1% of market cap per month. It could depend on your strategy too.
August 14th, 2008 at 5:58 pm
Last I looked – and it’s been a while – Joe Schmuckatelli could do 2:1 in almost any margin account and 4:1 in stocks as a “pattern day trader,” and the inherent leverage of stock index futures at a typical retail broker’s minimum margin was about 15:1, so I think to get 50:1 on stock index futures, you’d be explicitly borrowing money.
For the timeframes and issues I’m interested in – weeks or months, equities and ETFs – I think that if a trading system requires leverage to be worth executing, it’s not a very good system. As activity ramps up and timeframes ramp down, leverage makes more sense, but generally I’m talking about unleveraged or lightly leveraged – 2:1 at max – strategies.
From my small sampling of intraday (stock) data, there’s a lot more liquidity in the 15 minutes at open and close than any other 15 minute period during the day, if you think in terms of dollar value exchanged per unit time.
I’m trying to build the idea of maximum scalability from the bottom up, because I can see the number of stocks by characteristic that are held in test by particular systems, and need to infer the maximum AUM that can run through them.
Perhaps a better number (for my purposes) than the PRIMECAP fund’s median market cap would be their MINIMUM market cap, and the range of market cap that makes up the bottom decile of their holdings …
August 16th, 2008 at 9:29 am
“For the timeframes and issues I’m interested in – weeks or months, equities and ETFs – I think that if a trading system requires leverage to be worth executing, it’s not a very good system. As activity ramps up and timeframes ramp down, leverage makes more sense, but generally I’m talking about unleveraged or lightly leveraged – 2:1 at max – strategies.”
Well, there’s a good body of emerging research on risk-parity portfolios that shows you can employ leverage to produce returns at a lower volatility vs. unlevered returns.
I’m just not good enough at math to put one together :) - here’ s an example/discussion:
http://dl.getdropbox.com/u/17693/2006%20AIC%20-%20Qian%2C%20Edward.pdf
August 16th, 2008 at 9:46 am
Risk-parity, beta risk, covariance matrix? That’s not research, that’s math-turbation. They’ll find out - eventually, at the extremes - that their leverage produces a HIGHER volatility of returns over the long run.
If it sounds like bullshit that’s too good to be true (”use leverage for higher returns at lower volatility”), then it IS bullshit that’s too good to be true.
August 16th, 2008 at 10:48 am
LOL Bill - I’m not planning on using it - just curious if you had looked at it. Clearly, the answer is yes!
Well, I guess I can put that area of research to bed….