2 ms·
i just want to check my understanding of the algorithmic trading "world", so please do jump in. Once upon a time (1986ish) the equities and bond trading world
by lifeisstillgood 10y ago
i just want to check my understanding of the algorithmic trading "world", so please do jump in.
Once upon a time (1986ish) the equities and bond trading world was run by humans talking to humans and agreeing deals, the prices then fed into computer systems and the exchanges passed the prices around to make things mostly fair.
Fair of course is relative, the Eco-system was very hierarchical, with major institutions at the top, trading between each other at low fees, with brokers feeding up into them and retail shops feeding into major brokers. The customer got a raw deal, being charged heavy fees per transaction, and getting a poor "spread".
Spread was where the major institutions made their money. Human traders effectively bought very low and sold very high - both because they were human and could
not easily handle algorithms in their heads and because who was going to stop them? At the top of the hierarchy traders got to see both sides of every trade - they could net trades off one against the other to make deals with little risk. And if it was not visible in a fair exchange they had even more leverage.
Spreadsheets took off around now, making it possible for one trader to plan and monitor his trades and look really good to his boss.
And then it became obvious that having a human in the spreadsheet-to-trade loop was sub optimal. A human with a spreadsheet still needed to dial a phone, make a decision, go to the toilet. A perl script could out perform him.
And at the time the algorithms were simple. If Exxon's share price dropped then pretty obviously other oil companies would drop too, but so would say car company stocks, but maybe coal miner shares would go up. And that's just in LSE - the same goes for Hong Kong and Chicago. Those correlations I could work out in a perl script. (OK, 1980, maybe some Basic :-)
And so algo trading was feasible with really tiny hardware - because the correlations in the world markets were simple, and large. And so low latency trading started. Because if I can use my ZX spectrum of my Commodore 64 to beat major traders to the punch, then all you need is a faster computer than the commodore and you beat me to the punch. And so it goes.
Fast forward twenty years and
- the hierarchy of the past is mostly still in place. Retail shops pull in the customers money, pass it upwards to brokers and they deal with traders at large banks. However the traders are much reduced, the volumes they do are orders of magnitude larger now.
- the spread has gone. Major institutions make money on tiny margins and tiny fees and just do vast vast volumes. Major FX desks will make maybe 10 USD on a billion dollars of Eurodollar trades (I think).
- the spread has gone for the algo traders. The reason PhD's are needed is because the correlations and arbitrage is all eaten up. The wins are few and far between and mostly need real world events (Brexit)
- this is generally good, there is more trade on open exchanges (good for everyone) there is smaller spreads (good for customers). The break neck automation to a good for contractors like me :-)
I'm not sure where I am going with this to be honest - but mostly it's that I am sure zip line is a good library, that the core part is written in the way a proprietary engine would look if someone took a year to rewrite it, but the core tech will not give you any edge - that edge has gone. The correlations have gone except in esoteric areas.
If you want the edge, you need to be at the top of the tree again.