Define 'strategy'.

What I'm wondering is, how many robust automated trading strategies would an average or even small size quant fund have in operation at any one time? 2, 5, 10?

Also when people say they have a particular %age return with a particular strategy or even fund, what size account are they talking about?

Lets say you have an average profit of $1000 per day. This could be 10% per day on a $10,000 account/fund or the same return could be 0.1% per day on a $1,000,000 account/fund. People never seem to relate the account size to the return. I would have thought the two must be stated in order for the numbers to be relevant...?

The limits of question depends on sample time of data, processing time of that data, and execution time. That would accumulate across the strategies until you run out of processing power. How each person/firm decides to load their processing system is individual.

Noah,

"...how many robust automated trading strategies would an average or even small size quant fund have in operation at any one time?"

If you need an estimate or a result of survey as a direct answer on your question, then stop reading because I do not know.

The term 'trading strategy' is used at least in two senses. The first is a list of transactions (1, -2, 0, 1) meaning buy one, sell two, do nothing, and buy one (contracts or share units). Given a corresponding list of prices (150, 200, 170, 130) and costs (-10, -10, -10, -10) one comes to a profit and loss, PL, value -10 + 50 - 10 -10 + 70 - 10 = 80.

The second sense is a set of rules, which leads to the list of transactions (1, -2, 0, 1).

Consider two neighboring time points and assume that the long or short position is limited to one contract. If the position was zero at the beginning, then the following strategies (lists of actions) in the first sense are possible (-1, 0), (-1, 1), (-1, 2), (0, -1), (0, 0), (0, 1), (1, -1), (1, 0), (1, -2). For instance, (1, 1) would create position 2 and is out of the list.

Clearly, that for any finite discrete list of prices where reinvesting is not applied and the position size is restricted to -1, 0, or 1. The number of strategies in the first sense is finite. It can be determined using combinatory laws. While the number of strategies in the second sense is countable (strictly speaking infinite), they still realize only one of the strategies of the first kind. Simply, many sets of rules after some point will result in one of the strategies in the first sense.

If one applies mark-to-market evaluation and transactions cost is not zero, then (0, -1), (0, 1) are losses. The (0, 0) always exist. It is a do nothing strategy. The remaining cannot be evaluated as a definite profit or loss without knowledge of prices and costs.

"People never seem to relate the account size to the return"

When a strategy is evaluated not only absolute profit or loss are used for judging about it. Return on investment, margin, account, duration of the trade, and many other characteristics are ordinary considered. Many of them do include profits or losses as key elements of the entire measure.

Best Regards,

Valerii

Hi Noah: Note that often a very, very, very large quant shop won't need large returns because of leverage. Also, sometimes a similar strategy can be used say on fx and equities so defining the # of strategies is difficult.

Also, It will be very difficult to obtain the actual returns of quant shops-hedge funds except for the ones that perform well so you'll have a biased sample if you try to go that route.

Ah lads, good God, why the hell are ye making life sooo complicated?

Ok fair enough, if the one strategy can be run on currencies or futures or comms or whatever then lets call that just that a single strategy. The question is simple. Let me put this a different way...... how many strategies does it take to make a good fund??

Thanks Mark. Anyone else get what I mean??

Okay I'll bite. I used to work for a large (>>$10bn) managed futures shop. There were three main fund styles. In the main and second largest fund on each asset class ran a slightly different set of models, but there was a lot of overlap so lets not count those. Let's also not count multiple variations of the same idea, eg different moving average speeds. So we're perhaps talking about maybe 10-20 "strategies". There was then another fund which was multi-strategy, so deliberately contained a lot of stuff - perhaps another 20 or so 'strategies'.

"How many strategies does it take to make a good fund?"

Now if I was running a small, single style, managed futures shop I'd probably only use half a dozen or so strategies (roughly what I use in my current trading). You get diminishing returns from additional strategies, and a lot of the fancier things need non price data to work which means more cost and effort. In a larger fund you can justify the manpower you need to throw at something like that to generate an extra few bp. In a smaller fund no, and most (85%) of the performance would come from the first two strategies.

On the other hand a small equity neutral fund could easily have 20 factors - there's less work for generating each additional factor assuming they are all coming from the same raw accounting and price numbers - although again there is diminishing returns beyond a certain point.

Does that help?

Noah, when you comment

"Lets say you have an average profit of $1000 per day. This could be 10% per day on a $10,000 account/fund or the same return could be 0.1% per day on a $1,000,000 account/fund. People never seem to relate the account size to the return. I would have thought the two must be stated in order for the numbers to be relevant...? "

you are looking for the following ratio:

Profit/DrawDown

This ratio is constant no matter the size of the fund.

Indeed to mention a %X return without any further explanations is meaningless.

@Noah, your question is totally disconnected! You present as example: <...Lets say you have an average profit of $1000 per day > and then talk in term of <... a decent quant> fund. No decent fund manager would even consider such a premise. It would be like playing with and for peanuts.

Why a hedge fund might use different trading strategies is related to how much they know of the future about their trading strategies. For instance, operating on the same data set (same stock selection), the fund could at all times order its trading strategies by productivity which is the same as asking the question: is

.

Σ(H(S#n).*ΔP) >? Σ(H(S#3).*ΔP) >? Σ(H(S#2).*ΔP) >? Σ(H(S#1).*ΔP) >? Σ(H(B&H).*ΔP) ?

If you have a trading strategy that greatly supersedes all others, then why trade anything else? In such a case then S#n would be selected. If they are all about the same, meaning producing about the same output, then you could spread the capital to a selected few of what you consider your best trading strategies. You would do this only because you don't know before hand which would come out on top. But this often condemns you to the average performance of the selected group of strategies. If you take your 10 best trading strategies, they still all have to be funded. And since you started with an objective of $1000/day, your discourse starts to be meaningless (making $100 per strategy per day!).

So to your other question: <...why the hell are ye making life sooo complicated? > Well, maybe because it is. May I suggest looking more closely at the math of your stated problem.

I'm going to take a slightly different tack in my answer to this question.

As far as a traditional quant hedge fund is concerned I would say that the need for additional strategies relates, at least in part, to diminishing rates of return in the existing strategies as the fund scales up. So, for example, a single strategy might suffice for the first $250M - $500M, but then a second strategy would be required to take the fund to north of $1Bn in AUM, etc.

In the HFT world it often works very differently. For example, I am currently running 14 variations of the same high frequency spread strategy, with different profit targets, stop losses and entry thresholds. Why? Well one reason is that HFT strategies can be so latency sensitive that you might easily miss profit opportunities by running only a single instance; by running several of them you are diversifying across time and ensuring the possibility of multiple entry points. One tends to do this in HFT, rather than try to "optimize" a set of strategy parameters as you might in a low frequency context, because the chances that the configuration selected is really optimal is very low, giving the sensitivity to initial conditions.

42.

Noah,

I agree with Jonathan post on scalability of strategies.

We had three different FX strategies.

One was a long term trend model (very low trading frequency) and two were medium frequency styles of an entirely different nature so that the correlation between the strategies was pretty low.

kind regards

Marc

IMHO -

The simple answer is that the number of strategies is likely a factor of the AUM, and most recent 3 year returns for the underlying AM, or HF. Shops that focus on equities will most likely run a much larger number of strategies compared to Forex, Fixed Income, Futures, Options, although options for any asset class may catch up quickly within the next 5 years.

Shops with larger AUM also have more varied and diversified HR; quants, traders, analysts, & product lines, so that there will be more idea generation and risk taking, and algos fired up.

Expensive institutional level EMS and OMS, along with co-location and other factors will also limit the number of players that can scale up into VL # of strategies. But, once in the stratosphere, complex event processing and other big data solutions begin to come into play, again very expensive.

From direct knowledge, at least here in Brazil, AUM does play a role, but 5-15 strategies seems to be the sweet spot. As AUM grows it's not a matter of increasing the # of strategies, but rather pointing your working strategies at a larger universe of symbols.

Thanks to everyone who took the time to contribute to this thread. Some of the comments have been especially interesting and indeed helpful.

It seems that you might not necessarily need dozens of proven strategies in order to drive a decent fund and especially so if the one or two or three strategies can be replicated across different timeframes and symbols/instruments and by simply making very small adjustments to entries and profit targets etc to prevent them from converging on one another and thus biting into the same liquidity by all triggering on the same signal.

Robert that helps a lot. I am in the process of trying to add some diversity to my current set up and it always helps to get an idea of what others think. I am currently running some pretty good short term strategies but I also need to try to get some medium term ones up and running.

Marc thanks for that. When you say medium term, what level of frequency of entries have you in mind? While we're on it, what number of set-up entries would you rate as HFT?

Matthew I appreciate your input. Yes it seems the natural progression is as the AUM gets bigger so does the capability/necessity to increase the number of strategies to keep up with the return expectations. And of course, yes collocation is a must in order to get the required reliability and uptime. I am based in Ireland but have everything colocateded to Chicago and hope to expand it into NY and Frankfurt in 2015.

Thanks again to everyone thus far.........oh, and Happy New Year to all..!

Jonathan Kinlay answer is good, gets my vote. One can scale a single strategy to maybe $1B (10-20% gross alpha ret). Or even more in lower Sharpe ratio stuff (say ~1SR) where a single take on LS Equity or Quant Macro might even be 5-10B (10-20% gross alpha ret). Of course, the underlying realities are more of a continuum where BC those market universes are wide, one might see different "sets" of strategies based on diff securities, markets, industries, timeframes, etc under the hood. But in other cases maybe not. Overall, asking a 1-dimensional kinda naive question like this doesn't really get you far. Look at fund firms and ask how many each one sees themselves as running, and why they see things that way, assets, ret perf to each strategies, etc. (Maybe a 100 more). Some data before theory. OR... Ask what the ideal strategy # might be, and how might one work towards that? (theoretical question... In "ideal" context). MSJ

Lets try to define "Strategy" first. It is the setup of automated trading rules for a certain asset class, or cross asset classes that trigger BUY, SELL or HOLD signals for a specific group of securities, or cash, at any time to achieve Alpha. Ok?

Now, lets asume for simplicity sake (it was a simple question) that we have Alpha Strategies that operate in LFT and HFT

Capacity in one end and frecuency on the other define the number of transactions when trading AUM asuming operational capacity accordingly

Lets talk about LFT. Latency probably is not important, but Capacity is. Capacity relates to the feasibility of a certain strategy to produce alpha consistently trough different sizes of AUM. If AUM tends to grow alpha will tend to decrease, hence the need to add strategies to maintain alpha. Execution tactics might be needed in order to fill large orders in succesive days for a single trigger and operational concerns will unveil the need to add staff or not. It is also important to consider the lifespan of each strategy, a mature strategy properly back and forward tested will not be subject to mainteinance, it will be simply replaced upon lifespan end, defined as the lack to produce alpha, so it is also important to consider Quant effort to continuously create new strategies as markets evolve. My take is that around one or two dozen strategies will consistently produce alpha for each 250M USD AUM every 3 to 8 months in LFT, regardless of the Asset Class.

If the fund operates a small scale AUM, lets say, 5 to 100M USD probably one to five strategies will do.

In HFT we have a different problem: liquidity, instead of scale, is the issue, so a swarm of strategies of similar parametrization will need to be fully automated with a direct FIX connection to the Broker/Bourse at ultra low latency to be competitive. HFT algos tend to be very specific in nature. Prop Trading for WS Big Dogs for example might use around 10 HFT Algos for each billion traded on the prop desk daily

...