Hello Oskar,

There is some discussion of the topic on http://en.wikipedia.org/wiki/Technical_analysis, which includes references for further reading. I haven't read it, but one that may be enlightening is http://www.farmdoc.illinois.edu/marketing/agmas/reports/04_04/AgMAS04_04.pdf.

One way to approach the problem is to see if you can write an algorithm that trades only in SPY (or perhaps SPY and SH), and see if you can do better than a simple buy-and-hold in SPY. I've attached one of my attempts.

Grant

We can certainly write algorithms that will beat the S&P 500 in our backtests, and maybe even in paper trading!

I am trying to determine if anyone on this site has actually used algorithmic trading to turn a consistent profit. You would think that if there was some 'Black Box' that could turn consistent profits, large institutional players would be using them, but as far as I know they predominantly use fundamental analysis.

I'm hoping someone can provide evidence to the contrary!

Hello Oskar,

This is not an answer - far from it - but it is curious. It's a current job in London which I assume is more quantitative than fundamental. What struck me is it's the first time I've seem a Sharpe Ratio as part of a job spec.

"Job: High Frequency EFT Trader - leading prop trading firm
Industry beating hedge fund in the HFT space wants an established quant trader (at least 2 years of PNL history) to build a new business unit. Infrastructure in place for high frequency arbitrage on US and EU equities (trade life measured in microseconds).
Potential for building on existing infrastructure as needs require.
Candidate
Realised Sharpe Ratio greater than 3 over a two year period
MSc/PhD in a numerate discipline (Maths/Stats/Physics/Computer Science/Engineering)
Competent in at least one major programming language – Java/C++
Scope for operating as an independent trader, preferably you will want to build a team around you
Company
Award winning hedge fund with significant track record within HFT
Massive investments in technology - dedicated tech teams, exploring microwave technology
Membership of all major exchanges
No internal competition between trading groups
Shared services that can be used by any trading group on offer – quantitative researchers, back-testing facilities (with live trading data recorded over a decade), dedicated exchange connectivity teams, and more
Ownership of intellectual property on offer
Extremely competitive pay-out structure, % can increase with PNL figures"

As an I.T. contractor for nearly two decades I am wholly contemptuous of recruitment "consultants". This one seems to be fairly typical in that (s)he doesn't know the difference between EFT and an ETF. It's all money, I suppose....

P.

Peter, thanks for the reply. HFT is one area where algorithms can certainly produce profit, but that's not something that I, as an individual investor, am capable of competing with. My goal is to invest my own money (perhaps with some leverage) with an algorithm run through Quantopian, and given that the maximum frequency that Quantopian trades at is once per minute, HFT is out of the question! Even if it could trade faster, there is no way to compete with the big players that have trading machines colocated at the exchanges.

Therefore I am hoping that there exists an algorithm that holds positions for longer that can be consistently profitable! And if there is, wouldn't large investors be using it?

Hello Oskar,

I suppose it's a question of (biased) evolution. When retail investors first had access to the market via telephone to their brokers they only had access to half of the market i.e. they couldn't short and were no threat to the institutions. Then computers came along - I saw a webinar the other day where they said Thomas Peterffy (IB) started automated trading on NASDAQ in the 1980s and they asked him to stop. Later when retail investors do have electronic access access to both sides of the market the institutions start their algo trading. When retail goes algo this becomes a FLOPS and/or data-mining arms race. When retail I.T. goes parallel and (cloud) scaleable the institutions go co-located and dark pools. I read the other day that the LSE will offer institutions with enough cash wireless access, presumably because there's no co-lo space left.

I'm starting to feel that the retail investor can't compete. The institutions will create their edge, illegally if they need to, and keep on going. Look JP Morgan's £5.1 Billion fine this week in the context of this:

http://www.huffingtonpost.com/2013/03/06/eric-holder-banks-too-big_n_2821741.html

It's a long time since Bobby Kennedy as Attorney General was committed to fighting organised crime.

P.

Hello Oskar,

You might try to quantify your requirements. When you say "beat the S&P 500" what does that mean? And "consistently"--are you thinking for the next year, or 40 years? How will you assess risk? Also, how much capital will be committed?

My sense is that it will be very challenging to construct a single active approach that will beat the S&P 500 over the long haul, taking expenses into account. In other words, you write the algo today, fund it, and when you check back in 40 years you will have done significantly poorer than if you'd just put the money in an S&P 500 ETF. But I could be wrong.

Grant

Grant,

I would say that the algorithm should beat the S&P 500 by at least 5% per year in returns until there is an 'evolution' in the way orders are placed as Peter summarized. In terms of acceptable risk, I think the Sharpe ratio should be at least 2 and maximum drawdown of 10%.

I agree that there is little hope for an algo written today to perform well for the next 40 years untouched (unless perhaps it is really good at learning), but I would hope there is an algo that would work for at least the next 5 years with minimum maintenance (unless there is an evolution).

My central question remains: Is there a reason why major players don't seem to be utilizing 'black boxes' to do their trading?

Hello Oskar,

I don't understand your central question regarding the use of 'black boxes' by major players. I don't have a background in the industry, but I suspect that the vast majority of major players (e.g. mutual/index funds, etfs, hedge funds, etc.) are largely automated (with varying degrees of human intervention). Nobody has rooms full of analysts in green eyeshades trying to sort out the next trade.

Given that your benchmark is the S&P 500, your requirement of a max. drawdown of 10% may be unrealistic. Just glancing at the S&P 500 over a long time period, my sense is that the probability of avoiding a >10% correction over a 5 year span is low. In other words, if your "algorithm" were simply a buy-and-hold of SPY, you'd likely experience a >10% drawdown over 5 years.

Grant

There are (real) quant shops which are NOT HFT who require backtest sharpe over 3.0 to look at your strategy, and this means production sharpe of around 2.0. So, yes, it is quite possible to beat the markets doing stat arb. How much volume can these strategies support is entirely another matter

Thanks Igor,

Do you have a specific stat arb strategy to share that would serve as an existence proof?

Grant