Arguments against Automated Trading

14 comments Written on July 5th, 2008 by
Categories: TradingSystems
I've been speaking with a few people lately and also looking at the comments on trading systems. A number of people feel that this concept of an "automated trading system" is very strange and it simply won't work.

Most people quote that they have seen a thousand of them earlier and they didn't work. While I understand systems tend to fail (especially after they are public for a while), I am willing to bet that most systems "appear" to fail because of the lack of discipline. Most people fear for their life after a few bad trades and have no real way to prove that a system has indeed stopped working; so they throw out the system.

If there's a series of losses the position sizing techniques (always only risk a percentage of equity, post losses) should keep the position down - but a discretionary impact may be to stop real money and paper trade the next few trades, until there is confirmation that the system has gone bust. Or, to put lesser money in. The concept of throwing out a system on a few trades is probably the most attractive to a human mind, yet, the logical thing is to await some quantitative proof that the system no longer works.

The other problem people have is they are incredulous. How can a system that is computerised and requires no human intervention, perform at any consistent rate? But that's precisely why it does, because it has no human emotion to contend with. The more humans get involved in trades, the more likely it is for mistakes to happen. Now there are many traders who are excellent at intuition, and the vast majority of trades happen on "gut feel" - so there is always the feeling of a loss of control when you let it all happen in an automated way. But you can't back-test "gut feel" and you can't establish how something might have worked in the past when it's based on intuition.

This goes for a lot of fundamental stuff too. People say that's better because it's a more logical way to invest. I doubt it. Fundamentals are scary in these times when companies will lie through their teeth to prevent their stock from falling, or scream buyback, or ask for SEBI to investigate - anything to protect their stock price. You can't trust anything anyone says - at least one thing you can trust is the stock price! And the concept of buy and hold will soon be apparently flawed, as people who bought RNRL at 212 and DLF at 1200 realise how interesting life becomes when you buy and hold for a long time.

Still, value works - and can be backtested - if there is a measure for value. You can check things like how far below we are from a long term moving average, how much cash a stock has in comparison to price, how much of past earnings growth over say 5 years is reflected in the P/E. These concepts can be back tested, and even automatically traded (imagine, automated fundamental trading!) The concept there may yield far too few trades - maybe a few a year - but may be far more attractive in the long term (10+ years)

Lastly people are cynical about such systems. They think it's a fad, it will go away. Yes, sorta like fundamental investing "went away". Or stock markets "go away". There are examples of HUGE profits from such systems - look at DE Shaw, Jim Simons of Rennaissance Technologies, Ken Griffin of Citadel or Man Investments. These guys manage billions and billions of dollars, through automated systems, and make a ton of money - some put their annual income (performance fees + management fees) as upwards of $1.5 billion. That's enough to silence most critics - because you only need one such case to DISPROVE the argument that systems don't work.

I'm not saying other forms of investing don't work. Warren Buffett's has, obviously, and so have those of Peter Lynch or Rakesh Jhunjhunwala. Or even the discretionary trading of John Arnold or Steve Cohen. But that doesn't mean pure automated systems don't work - they do, as the performance of the big hedge funds mentioned above are well known.

My aim, and Kaushik's, is to become that kind of automated trading firm for the Indian markets. Obviously we lack the money (otherwise life would be different) to invest in the technology and the connectivity/membership required to make this happen. Also the track record necessary to attract the right kind of money. Keeps us excited!

Related Posts Plugin for WordPress, Blogger...
About the Author:
http://www.capitalmind.in
The man behind Capital Mind. Deepak is a co-founder at MarketVision, a financial knowledge company. Deepak also provides data research and consulting services, and now lives in Bangalore. Connect with him at deepakshenoy@capitalmind.in.

14 comments “Arguments against Automated Trading”

>Deepak,
everyone is not easily adaptable to the change. Even I feel at times of being resisting to the change occuring around myself. So in that sense the ATS idea will/will not be very well received. But should it affect your aims’s? As long as you enjoy the profits even when the markets are high volatility, this should be good track record. Now do you need to prove this to everybody in general? IMHO, I dont think so.

And for the track record, it can only be achieved by putting REAL money over the period. Can there be any short cut for that? No.

Say, your system has enough STEAM and the investor has enough GUTS, thats it then. Even a few wealthy investors migth be more than enough!

Now say ATS takes off and achieves a good track record. How much ‘Public’ you want to make it! How much investment is perfect/ideal for the system to perform to optimum efficiency. Too less money, too much money in, hows does this go with the system?

My two pence.
As always Good Luck.

>Deepak,

“Most people quote that they have seen a thousand of them earlier and they didn’t work.”…imagine how many business plans that VC guys see, i’m sure millions and most never work out or get funded.

The whole “fundamental” research business is multi-billion dollar business and you are basically calling their baby ugly…so people will react and say ATS doesn’t work. But within most prop books is an ATS group trying for the holy grail.

>Siddharth: good point, there’s no need for formal acceptance, in fact the more secret the better! Must think about that. Thanks!

Manish: Good points there! NOt sure if I understand the VC angle – even we throw out more systems than we use (by a factor of 10!) but that doesn’t mean the whole concept doesn’t work.

Interesting about the prop books. ATS is already being used extensively? Do you have any sources I could look at? Very curious…

>Hello Deepak,
I would like to point out a few things I believe you should think on:
a) Automated trading system is an interesting concept. Now what happens if there’s a system failure during transaction execution? You may turn back and tell me “Hey Arpan, there’s only 5% chance of Windows rebooting and ISP hanging” but that in itself is a scary thought. Systems’ reliability is a big deal.
b) I am assuming out here that your broker has allowed you API access (http post/get). You might want to discuss the security features for the naysayers. Obviously though, if someone sends login/password info over get then God save him. In the US systems are time tested, in India we don’t know how secure this might be. Also how fast are these broker systems? Your entire game plan is dependent on the speed of interaction.
c) Strategies you take for risk control is something you might want to discuss at a high level. Having been an avid reader of your blog for the better part of my investing life, I have seen you doing things like cash flow analysis etc of stocks when you pick those at a fundamental level. I would like to see you maybe devote some time delving on the nuances of risk control in short term trades for us to ponder about. Just keeping stop losses might not be fine — you still lose, and if your system shows a bunch of false negatives on a single day you collectively lose a lot. That ain’t risk control.

Please note that my point of view is essentially what a technical person would like to tell his investors’ about the system.

–Arpan

>Arpan:
a) Automated trading is NOT the same as automated execution. I don’t care about execution if I generate two trades a month, for instance. I can literally call them in if I like. If trades are more frequent I can even hire a jobber to enter the trades as they come. Right now we get less than 10 trades a day even on our most frequent systems, and none that are highly sensitive to IMMEDIATE entry.

Right now we aren’t too bothered about auto-execution. It’s the trades themselves i.e. what to buy, when to buy, how much to buy and the sell side questions that are important.

While system reliability is important, it’s not ultra critical right now – we have faced transaction execution issues already, and are still working those bits out. When we have enough money we’ll have a more reliable set up (to ensure nothing at our end fails) – but right now we are using systems that don’t need ultra high reliability – can’t afford it.

b) We don’t care about speed, these are not high freq systems. When we need those we will set up the correct infrastructure.

c) There are hajaar elements of risk control – stop losses, risking only a small % of equity, finding out risk by doing tests on atr/volume/relative moves etc. I will talk about them as we go along. There are other elements of risk in back testing – MAE, drawdowns, number of continuous negative trades etc.

But good points. Eventually these will be of importance. Trading systems are about high quality data, back tested algorithms and very reliable execution. As you can see the algo part is only 1/3rd of the picture.

>Deepak,

Your comment in the original post “…Most people quote that they have seen a thousand of them earlier and they didn’t work..” A million billion more will be created and many will fail just like any business.

My comment about the VC space wasn’t clear. What I meant is that the entire VC industry space sees millions of ideas but only funds a tiny fraction. So does that mean the VC industry doesn’t work…clearly not.

Arpan,

I believe you are talking about two things:
1. Direct Market Access
2. Algo trading

If you combine the two, then you have a DE Shaw or Renaissance.

Designing a system to avoid system failure is like any technology structure. Working towards 99.999% of uptime, which equates to 5 minutes a year of down time. In India, with the advent of DMA, at least weak link is elimiated – the human dealer who could be on a tea break when an order is entered. To eliminate other issues you could have your system hosted on a brokers platform and since it’s a black box, theoretically can run automated.

The API you are talking about is available, most brokers around the world talk FIX, and that has been around for ages and been used quite successfully.

>I think then automated trading system is a definite misnomer. You should call it like ‘an automated data analyzer for trading suggestions’ or so. However one thing is really unclear — you are going after very short term trends. The system is analyzing data after every couple of ticks. You say that you would do manual execution. By the time you actually execute what happens if the trends change? Fair question, right if you are gunning for ultra quick moves in a volatile market?

–Arpan

>Arpan: Auto trading simply means no human discretion in the tactics. It may mean execution, but automated trading need not include automated execution. Like Manish said, DMA will solve that problem by providing an API.

To be honest, I won’t be talking more on exactly what we are using, but suffice it to say that if we need tick by tick execution, we will invest in systems that will perform as fast….

>By the way, have you back tested Graham Greenblatts’s method as suggested in “The little book that beats the market”. In fact it is also mentioned in the “Dhando Investor” by Mohnish Pabrai. The only problem I find is to find accurate and updated filters to screen the stocks. I ahve been trying this for the last six months (I know …the sample time is too small… I am continuing …). I would like to know if you have done any work on it and any suggestions to screen the stocks based on RONW or ROCE and earning yield??. If not, would you consider building one albeit for payment (unlike the one available at Magicformulainvesting.com)

>Deepak,

I have been intrigued by your posts recently on your automated trading systems and have been reading up on them. However, with my almost non existent basic fundas, I am floundering a bit. I do have a few questions though.

1. You (and others as well) mention that the systems will eventually fail. Why does this happen?

2. If we assume that the systems do fail, how do we know that whatever backtesting that you have done are applicable in real life if these are not applicable any more. I am not sure if I am clear on this one, but what I am trying to ask is this.
a. Successful Systems fail over time
b. To test a system you need to try it over past historical data.
v. What worked earlier need not work now (see point 1).
I think it basically depends on how long the window for a system to work is.

3. How do you figure out that the system has started to fail till you hit the losses? My question is on the lines of while the systems do not have human emotions that interfere with the trades, how would you react if your system (lemme call it your second baby) starts acting weird?

4. Lastly is this not susceptible to diminishing returns?

Regards,
Kiran

>Kiran: Firstly thanks for the questions, very valid points.

1) Systems fail for many reasons, some of them include changing times, or that too many people are using the same kind of system or what not. It may not happen often, and sometimes tweaking it a little bit may bring back performance. The systems used by the legendary Turtles for instance have faded because too many people used them – yet, even today people use variations of them.

2) The bet is that you are in the better part of the window. Thus it’s not just important to create a system, but also to test it with real money to see if results are consistent. Remember, most good systems last for 5-10 years or more, and we in India are only starting to see enough liquidity for system trading.

3) There has to be a test for a system failure. We have many thoughts on this, including checking statistical deviation of P&L or equity curves on the current run, and seeing if certain patterns violate back tests – historical drawdowns, no of continuous unprofitable trades etc. It’s a matter of emotion here but over time we’ll set up formulas for ditching systems too (haven’t eyt gotten that far)

4) Im not sure it is – if too many people get in, sure, but I don’t expect that for a few years. Eventually returns will diminish as capital managed increases – you can’t get the same kind of return on 1000 cr. as you can on 10 cr.

>It certainly looks interesting, i wish you all the best.

From the tax angle, with a very small number of trades during an year, I have to keep the records upto date. Just curious to know how this is taken care of in automated trading systems.

>Dear Kiran and anonymous,

To elaborate on Deepak’s answers to a few questions which have been posed here-

System Reliability Problems and Automated Execution: One solution is to use a virtual hosting service like rackforce.com (no affiliation). This is good for a nascent fund that wants the flexibility to grow without investing in hardware upfront.

Systems Will Eventually Fail: One way to dramatically extend the profitable life of a strategy is to have it automatically re-optimize parameters based on continuously refreshed recent historic data. This is my specialty and it is one of the most common ways AI algorithms (neural nets, genetic algos, SVMs, etc) are integrated into a trading system. For example let’s say you are using a turtle strategy- AI would automatically choose the optimal lengths of time for each moving average and update them, say, once a week or so to fit the current trend.

Regards,
Max Dama

>Very interesting, Max. I am also looking at ways to figure out when systems have collapsed and how to backtest a retrofitting algo (i.e. back test the changing of parameters based on historical data, say every week or month) The results are extremely encouraging.