I very strongly believe that the way to do well in trading is to find an uncompetitive niche.
Most traders you meet really have a few specialty trades they put on regularly and milk.
Those are their uncompetitive niches they’ve found.
…
Starting small doesn’t mean staying small.
Many people will start off in smaller name assets and eventually grow to larger names.
Having an already profitable system but with limited capacity is a much better position to grow from compared to having nothing and trying to carve out a chunk of an established market.
…
That’s why you typically see the pattern of taker/taker -> taker/maker -> maker/maker which I touch on in this thread:
In trading, you only make money if you are one of the best at something.
You need to beat out everyone else. So you need to do something niche that lacks scale if you want a chance of getting to the top near the start of your career.
…
You won’t be anywhere near the top in the generally popular areas and the reward for that is 0 when you’re pursuing high return active strategies.
…
So not only is it the best starting point - it often is the only realistic place people stand a chance at competing in without huge teams.
…
That niche can be:
-by market
-by market cap
-strategy
…
You can trade specific markets such as crypto defi, and perhaps even sub areas within that.
…
Smaller cap assets tend to have more opportunities as well but less capacity.
…
Strategy wise, you can build strategy specific knowledge and master that strategy, becoming the best at it, but also concentrating risk in the belief that this strategy is the place to be for profits.
…
Anything that can scale will be competitive as fuck.
High liquidity/ capacity or generalisability is what creates these conditions.
…
Highly liquid markets don’t necessarily mean high capacity strategies.
For pairs trading you may have complex spreads where putting on size costs you too much
OR
Such high turnover that you can’t run much with the strategy.
…
BUT it’s a good rule of thumb generally to consider the liquidity of the asset pretty proportional to the capacity of the strategy.
…
Generalisability of the strategy is key as well.
Reacting to Elon Musks tweets about DOGE is not very scalable.
Reacting to Twitter accounts is more scalable.
A firm focusing on sentiment data is very scalable.
…
They also increase in competitiveness with how unscaleable what you are doing is.
…
You don’t need to start off dominating trading in the S&P500, you’ve got your whole career to work there and most traders if you read into it far enough make highly profitable careers off their own slices of the pie they’ve carved out regardless.
…
If you’re looking for ideas on niche, low capacity, strategies aimed at retail players instead of institutional players (that won’t bother to only run a couple mil tops) - then feel free to check out the blog www dot algos dot org
…
I cover multiple strategies like this in the small trader alpha series in great detail. These strategies include:
The statistical processes or models for simulating the book/fills are p. much useless. None of it is even slightly useful in indicating whether it'll work or not.
...
You can't really simulate limit fills for market making, as in you can't simulate trades that never happened in the real data.
There is some ability to do it with the historical trade feed, and it only works if your market isn't super competitive because you assume your placements in the book have no affect on the other orders in the book / the taker flow.
For shitcoin arbs, it's a okay-ish assumption.
So from there you simply take the historical trades, and given your placement as an offset from midprice in bps, you simulate how much you would've been filled.
Quants like to think of themselves as fully automated and machine-like in their work, but the truth is honestly far from that.
No part doesn't touch a human hand.
Don't believe me?
...
Implementation:
This is perhaps one of the most automated parts, but manual components are still critical to the process.
Most HFT firms have traders manually tune parameters. There's ways to do it automatically, but humans dig around when they spot something funny - algos don't.
The only time I’ve seen neural networks work in practice for quant trading.
A thread. 🧵
…
I’ll start with some caution. Just because I’ve never seen it, doesn’t mean I don’t know it exists. I know XTX and such have huge ML models, likely using neural networks or genetic algorithms, but I’ve never seen it with my own eyes so hence the subject of this thread.
…
It’s very interesting that most beginners believe complexity is the answer to their problems when it really is the opposite.
Simplicity rules the land.
The key to success with neural networks was regularization.