Cool project! There used to be a really useful startup in this space, TradeWave, that would allow users to deploy their own custom code onto their platform and run it live against multiple exchanges. It also supported backtesting (testing your algorithm against historical data). Unfortunately, it seems like they were unable to find product market fit and since a few years ago their domain has been down.
I often wonder about the philosophy of making a trading bot. On one hand, you think that maybe by throwing enough brain power at the problem you might be able to crack the magical nut and print money from thin air. On the other hand, you know that millions of other incredibly talented and skilled people are also working on the same problem, and no one has really had much success such as to be notable.
On the third hand, it seems like common practice to use algorithms on Wall Street and in hedge funds, with many people claiming that the entire stock market is actually just 90%+ bot activity. So clearly it's possible to make money with bots, right? Otherwise nobody important would keep using them? I know that response time also matters with big firms spending billions of dollars just to be centimeters closer to the NYSE. But you can't just get rich from being faster than the next guy, can you? Don't you still have to be able to make all the right moves with the data coming in?
Maybe the secret trick is to become a hedge fund first, and then figure out how to make money later.
> On the other hand, you know that millions of other incredibly talented and skilled people are also working on the same problem, and no one has really had much success such as to be notable.
There have definitely been successes in this space, but they are always closed source within hedge funds and don’t usually offer up their algorithm for other people to use. Even internally in hedge funds they won’t share their algo trading strategies they have found, because if someone goes to another firm and replicates them then the advantage is lost. Usually these are based on understanding a specific signal that you can estimate a sectors value from.
As soon as others can use the algorithm any ability it has to beat the market is instantly lost.
To be honest it's not instant as everything has time lags and I'm assuming new strategies are very quickly incorporated into the market price, but if you have a market-beating algorithm and others start using it, that reduces the returns you generate potentially down to zero (depending on how efficient you think the market is).
The idea of an algorithm is that effectively you are wanting to work out the market value (or potential future value) of stocks/commodities to calculate if they are underpriced, i.e. you think a stock valued at £0.90 is actually worth £0.95, you buy them at £0.90 and hope to profit when people realise it is worth more. This gap between what you calculate, and the current market price, is the room you have to profit.
If others in the market start to use the same calculations as you, everyone will benefit from a more 'accurate' view of the underlying value of the asset, and this will change the market price. If you can't be the first-mover on this anymore you have lost your advantage.
Because of this, automated trading bots that actually work (which tend to be highly domain-specific, e.g. a bot that just estimates oil futures based on satellite imagery) tend to be sold to hedge/quant funds who can exploit the information asymmetry without disclosing the trading strategy.
> a bot that just estimates oil futures based on satellite imagery
Are there any resources containing lists of software projects or companies that do work like this? This and using satellite imagery to identify car models so you can estimate the average income of an area from its parking lots seem so fascinating to me. I imagine there's probably millions of unrelated, free data sources out there that could be linked together into something useful for market trading.
Although some novel strategies will be based on using publicly available data sources, a lot of it is often working out how to extract new data to create a private data source which has information others can’t access.
Examples of funds would be Renaissance Technologies and Two Sigma. It’s worth noting that these funds have really suffered during Covid as the change in the market has been too big for their models to support.
It’s not as instant as the commentator makes it out to be, or as decisive. You have weights of X and they have weights of Y. The capacity for the algorithm is $2mm of risk but you can only deploy 200k right now because your fund is smaller, etc.
> millions of other incredibly talented and skilled people are also working on the same problem, and no one has really had much success such as to be notable
I wonder how many strategies are profitable, but not at the scale that would be worthwhile for mid/large sized hedge funds to actually use. What I mean is that let's say I come up with some relatively naive strategy, that consistently delivers sub-1000/day profit. Can I just throw x1000 more monies at it and expect it to deliver 1M/day? I don't think these things scale linearly.
Just anecdotally from reading about this stuff online, it's quite hard to do know and very crowded. Apparently alpha (profits uncorrelated with general market movements) generating strategies rarely last more than a few months before they stop working. This is why a lot of the hedge funds in this space (de shaw, citadel) actually end up dedicating more to discretionary (human strategies) because quant/automated ones are saturated and can't accommodate large size.
But again this is what I've gathered from reading about this stuff, maybe someone with industry experience can chip in
No, you are looking for market inefficiencies, these have some size and it's improbable you'd find such a large one that nobody else found. And you'd start moving the market.
I often wonder about the philosophy of making a trading bot. On one hand, you think that maybe by throwing enough brain power at the problem you might be able to crack the magical nut and print money from thin air. On the other hand, you know that millions of other incredibly talented and skilled people are also working on the same problem, and no one has really had much success such as to be notable.
On the third hand, it seems like common practice to use algorithms on Wall Street and in hedge funds, with many people claiming that the entire stock market is actually just 90%+ bot activity. So clearly it's possible to make money with bots, right? Otherwise nobody important would keep using them? I know that response time also matters with big firms spending billions of dollars just to be centimeters closer to the NYSE. But you can't just get rich from being faster than the next guy, can you? Don't you still have to be able to make all the right moves with the data coming in?
Maybe the secret trick is to become a hedge fund first, and then figure out how to make money later.