i tell you a trading edge
i tell you a trading edge.
it'll probably look something like
x tends to get mispriced when y happens because z
ideally we have some causality we understand or strongly suspect
and some empirical evidence of this being the case, in the past, or in similar situations.
that's the initial job: finding some exploitable pricing distortion.
not much point doing anything else, until you have something like that.
but you still have to design and run a set of processes to exploit it.
there are a ton of different ways you'll fk that up.
what you're really trying to do is design a set of processes and systems that:
- get you exposed to the mispricing in the simplest way possible
- minimizes risk exposures you don't think you'll get paid for
- is manageable, robust, and has the minimum moving parts
the first one is never straightforward.
there's never a "correct" answer to it, because the market is complex and evolving, and you never KNOW what is really happening.
you will have:
- a concept of what is causing a pricing distortion
- mixed noisy evidence of that
so you gotta weigh up a bunch of conflicting data, and think about trade-offs, and try to make some sensible judgements on:
- what the effect really is and how it plays out
- the cleanest set of exposures to harness it
the most common mistake here is to be too confident about the future looking like that past.
ultimately, you're always going to be assuming it does, to an extent. (that's kinda the basis of everything we do.)...
... but markets are competitive and adapting constantly - so you need to be a bit paranoid.
the data doesn't help you infer cause -> effect anywhere near as much as you'd like...
there's a huge amount of variance in market outcomes cos a ton of things are happening all the time - so, to estimate things well, you'd like a lot of data.
but markets are adapting constantly: the newest data is the most valuable, and old data is becoming less relevant.
so you gotta combine your intuition with the data analysis and make some sensible decisions under huge amounts of uncertainty.
mostly this comes down to looking for things that are consistent or changes in behaviour that can be understood in terms of other things.
and ultimately you're going to have a punt on making some decisions on what the effect is and the best set of exposures to give you a good chance of harnessing it in a large set of uncertain states.
you're trying to throw shit in roughly the right direction.
you want simple and robust.
you do not want complicated or optimized heavily to uncertain states of the world.
you also need to get confident that the costs of harnessing the edge are less than the expected returns from it.
there are a lot of ppl who can trade faster and cheaper than you - so you need to be ruthless at minimizing costs.
and you also need to understand why somebody who can trade at less edge isn't trading it away.
understanding why a trade will work is mostly knowing why it sucks.
so, first, try to understand the effect and how to harness it...
the second thing you must do is work out is how to minimize the risks that you aren't rewarded for.
this is usually trivial at first glance..
but what you'll tend to find is that the cost of removing all other risks is higher than the expected returns from the trade itself.
so you have to be pragmatic.
what risks can accept? what can i not accept?
how else, through turnover or breadth can i reduce variance?
ultimately, being prepared to accept and manage risk others don't want can be an edge in itself.
you can go for opportunities others must avoid, or compete harder for them.
finally,you have to be able to run the processes in the real world.
managing even simple processes is hard work. everything fails. data arrives late or wrong.
everything that can go wrong will go wrong.
you want to design a simple robust set of process that are the essential to harness the effect well.
nice-to-have goes out of the window.
a feature that requires 4 data feeds is worse than one that only needs one, even if it explains more variance.
keep it simple.
ultimately, running trading processes involves a lot of work.
keep things as simple as possible.
and be paranoid that you don't really understand things, and design stuff to be as robust as possible to an uncertain future.
beep...boop.