Systematic Traders - How many systems do you run?

[savagemp5]

U dun need a robust system depending on the speed and complexity of your system

For real ops, u have namely a few like:

- Market data input sys => DB sys
- DB sys => algo sys
- algo sys may contain alot depending.
- algo sys => orders sys
- orders => feedback and clearing sys

Its more of how u wan to decouple a system that does everything (its posible) or to break them up for better processing

 

[sle]

I am just thinking - in an institutional environment, you have people doing a lot of stuff that you end up doing yourself if you are a one-man show. Just the system administration and DBA duties should be taking significant amount of your time, I am not even talking about juggling development/research with actual trading. So I wanted to hear if you came up with some organizational gimmicks to improve your life, especially guys that are gray-box traders and are not running automated systems.

 

[gmst]

Quote from sle:

I have an interesting question. The ones of you who run multiple systems in different asset classes all by yourself - how do you handle the IT/data and other organizational aspects? How many computers do you have? Do you run your own data DB and data collection?

I waited for others to respond, because I think my response will most likely sound less than attractive to you. I have started to believe that even with rudimentary tools its possible to cover a lot of ground.

Currently, I have just one laptop - thats all. All my trading, research, ET time, emails etc. happen on this 14.2 screen. This leads to huge waste of efficiency as I am sure I am wasting at least 30 minutes everyday moving from window to window and application to application. But, I keep travelling, laptop allows me that! I only trade futures and fx - so below is in that context.

I use following tools:
1) Multicharts - my tool#1 for strategy development on single markets. MC automatically saves top of the book data from IB into its database. Once you set up the connection with real time data from IB, it runs. So, zero time spent on DB after initial setup. If you want to use MC for strategy development at all, and you trade small number of markets, it will prove to be an ok DB tool. If you want to store maybe 1000 market data lines simultaneously and query, then I guess you would have to go with a more specialized DB. If you want to store whole book data, you can't do it in MC. If you want to store options data across strikes and markets, won't be possible again - imo, never tried it. MC won't do baskets, pairs, stat arb etc.

2) Excel - my tool#1 for data crunching, keeping stats, constructing portfolios of systems etc.

3) Matlab - Tool#1 for doing advanced multi-series data analysis and building econometric models (which is < 5% of my current approach). However, this will be my tool#1 for constructing a portfolio of systems with probably excel - that I am trying to get my head around in this thread.

4) I am an independent trader and operate out of my bedroom, so no organizational hassles, other than to keep reminding my family that this is serious business and they better treat it like that.

5) I do manual execution through IB TWS. I just don't have time to put in to ensure that auto execution will work uninterrupted. I am sure however that 2 yrs down the line, I would move as much futures trading as possible to auto-execution.

Now, all this might seem like a very basic setup of a retail trader indeed, but even these basic tools allow the development of a robust system development methodology, imo. Also, my edges are pretty good (imo). Sometime, over the next month, I will put very abridged results of my systems backtested performance. What I am trying to say is that even though having a bbg is cool, but it is possible to do without it especially if someone is starting out on his own and trading only futures/fx. Other basic thing I firmly believe in is that in trading, pareto principle applies to a more extreme degree. In a trading operation in a firm, only 5% of people create stuff of core value and it serves the other 95% people who are mostly support to earn a living. So, if guys from that 5% subset decide to launch their own fund, they really don't need a whole lot of staff to outperform 80% of CTAs out there. This is subjective and others may differ in their opinion here. But example exists where one man starts and goes on to build a trading super-firm. RenTec was started with just 2 people and Citadel by just 1 guy.

Not sure if any of this is relevant to your question, but its my approach and beliefs.

 

[gmst]

Quote from sle:

I am just thinking - in an institutional environment, you have people doing a lot of stuff that you end up doing yourself if you are a one-man show. Just the system administration and DBA duties should be taking significant amount of your time, I am not even talking about juggling development/research with actual trading. So I wanted to hear if you came up with some organizational gimmicks to improve your life, especially guys that are gray-box traders and are not running automated systems.

As an independent, you are responsible to do everything. Even though I like MC, I know if I had my own custom built trading platform with 3 programmers working on it. I would be able to get instant help on any operational issues - working with MC can take more than a day to resolve the issue sometimes. Also, with 3 programmers, you can get any functionality built up quickly in your trading system, compared to if you are using a canned software. So, yes as far as working is concerned, if someone is going independent after working for an institution, its a HUGE change. You become trader, system developer, programmer, DBA, modeler (e.g. writing vol models), researcher (reading academic papers), and of course salesman and face of your fund.

I will give you the distribution of my typical day. Working independently, on average I spend just 13% of my time for actual research, 20% time on internet trading forums and ET replying or reading historical posts (this time is similar in essence to what I would spend in an organization brainstorming ideas with colleagues and general chit-chat, just that signal to noise ratio is bad here), 20% time on following markets, 2% time resolving Multicharts issues, 20% actual trading and execution, 5% time programming, 5% reading and responding to emails, 10% time reading news.

Of course, I have different phases - like the research phase, when I would read academic papers and run stats for 50%+ of my working time. Programming phase - program for 50%+ of my time and scalping phase, watching markets 50%+ of time. Doesn't sound efficient use of time at all - but it is what it is.

 

[sle]

Quote from gmst:
So, if guys from that 5% subset decide to launch their own fund, they really don't need a whole lot of staff to outperform 80% of CTAs out there. This is subjective and others may differ in their opinion here. But example exists where one man starts and goes on to build a trading super-firm. RenTec was started with just 2 people and Citadel by just 1 guy.

Yeah, except in the end none of them actually started truly on their own. Usually, when an institutional guy with a senior pedigree starts a fund, he would get enough funding commitments day one to hire a few guys. Even with a smaller capital base, like ours (we are still under $50m), we are not starting alone - I have a partner who is more of a business development guy and we are looking to hire a junior trader right now and possibly a quant.

Anyway, I am very impressed by the fact that people here can actually do it all single-handedly - and it appears that some guys here run pretty complex strategy setups all on their own. Fixed costs is yet another bit - ok, so you are not renting an office, but you still have to buy and maintain your hardware, you still need to buy historical data, need to buy software, pay for an accountant etc. So even though it's way easier to extract alpha on a small capital base (making 100% of 200k with a Sharpe of 4 is way easier then making 10% on 100m with a Sharpe of 1), you have to be super-thrifty and smart about it, I guess.

 

[Mike805]

Quote from sle:

I have an interesting question. The ones of you who run multiple systems in different asset classes all by yourself - how do you handle the IT/data and other organizational aspects? How many computers do you have? Do you run your own data DB and data collection?

I have an approach that has grown organically over the last decade.

My current setup is about 18 systems that trade all liquid equities and futures. The goal is to have fundamentally opposing strategies that aren't correlated and I've done that to some extent, but, of course there are times when the approach fails. I've built custom tools to test all system signals for all products on a portfolio risk basis. Every system is weighted based on its ability to effect the portfolio from a drawdown perspective. This approach has already been discussed so I won't delve into it again. Its basically optimizing per system allocation and per system position allocation (this is based on custom metrics and volatility).

What I've found to be most beneficial is to separate out the functions into something logical:

1. Signal generator (signals can come from anywhere, but, they all have a standard format).

2. A gate-keeper (risk manager) function. Here all signals are parsed from a shared repository, checked for allocation, sized and sent for execution if risk checks permit.

3. A series of broker agnostic API interfaces. The gate-keeper tells any number of API's to send an order and the gate-keeper also requests all info back from the brokers.

So the risk manager is really the heart of the process and it allows me to easily interact with any data provider and any broker.

I use a shared directory (text files) for all generated signals so windows handles all directory IO, i.e. I can send signals across my network from any computer to the shared trade directory.

At the moment I use 2 machines but could probably get away with one. Its simple but I'm not HFT so its not necessary for me to get crazy about delay etc.

 

[Graper]

Quote from gmst:
Holy grail of trading is to have uncorrelated strategies on multiple markets. It will be interesting to know how many strategies do professional successful systematic traders run. So, if you are a successful systematic trader, your knowledge and ideas are welcome on this thread. If you are an aspiring systematic trader, please indicate so while posting. If you are a discretionary chap, stay away

Hi automated traders,
this is my first post in this amazing forum, I'm a small retail automated trader, and I'm trading forex with self-made trading systems since 2010.

Actually I trade three different strategies (Trend Following, Mean Reverting and Volatility Breakout type's) applied to the four most liquid currency pairs (eur/usd, gbp/usd,usd/chf,usdjpy).

All my strategies works intraday on a 1 hour timeframe.
So my portfolio counts 12 systems working toghether.

Not always I use all 12 systems, sometimes I have to stop a system if it cross above a certain level of drawdown, usually 10% in according to my risk appetite.
In that cases, before replace the system at the market, I wait for the system's equity retracement at the same level I've stopped its activity, before replace it at the market.

I've read many interesting theories in this thread about how to compose a balanced portfolio based on financial instrument correlation, and I'm in according with most of all.

From my point of view I think that don't really matter how many financial instruments you trade, but it's most important that you always trade every symbol that compose your portfolio with at least two different strategies.

For different strategies I mean strategies that make profits in different market conditions. The reason is quite simple, the financial istruments correlation level can changes during time, especially in actual globalized markets, conversely the correlation level betwen two or plus different strategies applied to the same instrument is always stable.

In this way the construction of really market neutral portfolios will be more efficient and durable. IMHO.

 

[CT10Gov]

Looks like I'm really late to this party - but isn't simulated max draw down of multiple normal random variates basically just a function of their covariance matrix? (though an ugly one with no analytical solution)?

So why is this any different from using portfolio variance (that you hate) x some factor?

Quote from gmst:

Continued from above.....especially the point 4 above - which is super complex imo.

In my view, the biggest challenge is - how to model combined DD for the portfolio. This is the real risk on portfolio level going forward. This is complicated because say you have 10 systems and you run 100 simulations (bare minimum imo) on each system to generate DD figures for each particular system. To know the portfolio DD from these individual Monte Carlos, you will need to do 100^10 = 100,00,00,00,00,00,00,00,00,00 calculations.

Doing even 10 simulations per system, would mean 10^10 = 10 billion calculations to compute portfolio DD. However, this simple approach (even though computationally time consuming) doesn't take into account correlations between the systems and hence is wrong!

The correct way to approach this problem, again imo, is to run a Monte Carlo on all the systems together, where you are sampling from a joint distribution using a correlation table. This will allow us to compute usable statistics for the portfolio DD etc. using 'just' 1000 simulations. Makes it computationally much easier to solve. The drawback in this case is that this assumes we know the 'true' correlations between systems and the correlations between systems are constant. We know in reality both these assumptions are wrong.

Thinking further - It seems like the "correct" way to backtest a portfolio of systems over multiple years would be to compute rolling correlations every year or every half year and use this rolling correlation table to run Monte Carlo on different systems using a joint distribution. Portfolio of systems will be re-balanced in this method every year or every half year.

To be continued....

 

[Sanz]

An interesting discussion...

I trade 3 strategies amongst 7 assets classes and 15 different instruments. Two are profit making strategies. The third is not meant to make a profit but rather decrease portfolio volatility.
I trade EOD only (at the open or the close) so infrastructure, wise, it's very simple. I use 1 amazon EC2 instance that runs the strategies EOD and e-mails me the daily orders. It takes me 5 minutes to type them in whether I am at home or traveling.

I have been trying to improve position sizing the strategies and it seems that performance based position sizing might not be the best solution. It seems that the least correlated strategy is very important in compounding profit (by minimizing portfolio volatility) even if it's performance is small or zero. So it should be allocated more than what it's performance implies. I have been looking at standard portfolio optimizing algos (risk parity, kelly, MPT, etc) as well as in forums like this one for ideas and guidance.


I coded a multi-strategy algo in Python in Quantopian to provide others with a template for multi-strategy testing. Feel free to try it or, even better, improve on it.
https://www.quantopian.com/posts/multi-strategy-example

Keep in mind I am not a programmer and this is my first time using Python.

 

[pepper_john]

If one has one system but with two different parameters, does it count as one or two systems?