171–180 of 254 posts
There's a cool article about this by Robert Carver who used to be a portfolio manager at one of the top quant funds.
https://qoppac.blogspot.co.uk/2015/11/david-versus-goliath.h...
I know a few people doing this, one person in particular who has discovered a near-zero-risk quirk that can be exploited with algorithms for fractional %age gain per transaction/cycle/ (kind of being vague here sorry) . He's been able to deploy large $MM amounts of capital to create significant gain nonetheless. However none of them will talk about it, certainly not on HN. I'm mentioning it simply to highlight I thin…
Earlier quoted context omitted.
Concur. You won't be able to beat players whose HFT systems are colocated in the same datacenters as the exchange. And you will not be able to beat those systems built using lower level languages like C++ or, worse yet, dedicated hardware. Source: 5+ years in HFT development
Why does a programming language matter in terms of algorithms? Can you elaborate on that?
This means that whoever is not the first to take that opportunity doesn't get it, and if you're reliably a millisecond slower than a competitor then you might as well not even try.
My question for everyone: Where do people get reliable data for back testing? Ideally I'd like data that goes back over 20 years. And with relatively few data integrity issues (e.g. does not exclude companies that no longer exist). Even more important: How do I know my data is accurate?
If you do equities Quantopian gives you a free backtesting platform with tick data going back to 2003. Pretty useful to just start hacking some ideas on it.
There are a few things to watch out for: 1. Systematic trading doesn't necessarily require an algorithm. For instance this rule might work (over a 5 year horizon, don't try it monthly): "buy very large cap stocks if their P/E goes below 4 and sell if it goes over 10." But you don't need an algo. 2. The market has long bull runs. So you might have an algo that has some long bias. It will seem to perform above chance.…
maybe I'm being naive but "buy very large cap stocks if their P/E goes below 4 and sell if it goes over 10." sounds kind of like an algorithm to me. Maybe it's just a ruleset?
If the entire stock market P/E goes below 4, you could use it. However when a single stock goes that low it implies that somebody knows something. Don't invest in a company with a P/E below 4 if they are going out of business. (the obvious example that probably never existed: a y2k company in 1999 - they are probably making a ton of money this year, but next year they will go out of business).
There are also "cyclical companies". They have a long history of having boom and bust years, it is well known that you buy when the P/E is high - the bust years when prices are low - and sell when the P/E gets low - those are the boom years when prices are high - but the boom will not last.
There are many other company specific things that can get in the way of any formula.
I hacked together my own scripted system that would arbitrage cryptocurrency across exchanges. It worked (for the most part), but it's been abandoned now. The best way I can think of to describe why is to say that while the low hanging fruit exists, there's far too little juice in it for it to be worth the squeeze. Others have explained that the problem they've encountered is counter-party risk in that some exchanges…
My arbitrage script was weighted to favor rebalancing my portfolio.
You probably can't do HFT trading because you need to have capital to reduce latency. Maybe you can rent servers very close to the trading centers, but this still will cost money. You can use http://www.quantopian.com to try out different algorithms. It will tell you how well your strategy works.
There was a great post on HN fairly recently written by someone who used to work in HFT. He talked about how they tapped the incoming network cable to read the incoming prices on an FPGA faster than they could make it through the OS's network stack. I think they were sending out trades in response to the new prices before they would have even made it to userspace on an OS.
Your time is an order of magnitude more expensive than that. Other expenses besides your time include making sure you have the fee structure to be competitive & getting a clearing partner that is ok with letting you fire off enough orders to make it worth while.
Our company works in the crypto space and we have a small research area that includes trading. We played with arbitrage strategies and have not seen a consistent return. Most times when you calculate a high return path it is because some exchange is not working really well (e.g. delaying transactions). Not saying that our observation is universal but I don't believe you can make right now a lot of money with arbitrage except in very discrete opportunities.
BTW, highly recommend the CCXT library[1] to connect with multiple exchanges.
I know a few people doing this, one person in particular who has discovered a near-zero-risk quirk that can be exploited with algorithms for fractional %age gain per transaction/cycle/ (kind of being vague here sorry) . He's been able to deploy large $MM amounts of capital to create significant gain nonetheless. However none of them will talk about it, certainly not on HN. I'm mentioning it simply to highlight I thin…
I had bigger plans for the project but lost interest after that.
Code is available if anyone is interested, though: https://github.com/nfriedly/Coin-Allocator