> the story of high-frequency trading is basically one of small smart firms undercutting big banks by being smarter and more automated and more efficient
Is that true? Isn't there a high barrier of entry? I was under the impression that large trading firms were building high-speed connections, which is obviously not something a small firm could ever do.
These days you can rent a co-located computer with direct connection to the exchange. The cost is a few grands per month. Not very cheap but definitely within reach of a small business. There are many small HF firms based all over the country that just rent 1 or 2 computers close to exchanges.
This is bad for big investment banks like Goldman because they no longer have a location advantage - you do not need an office in Manhattan to compete with big guys anymore.
Then my followup question would be: why do we actually need trading to be faster than the regular internet allows? For the objects being traded (companies) have time-constants that are far greater than the millisecond-range. And I hope the answer is not "because everybody else does it" :)
This can be partially fixed with a very technocratic market microstructure change (eliminating the subpenny rule). But politically that's very much a "huh?" point - imagine Bernie Sanders saying "I believe we should let traders quote in increments of 1/100 of a cent, not 1 cent".
Eliminating the sub-penny rule would probably be counter-productive for most US equities. You would not see further spread compression (most stocks' natural spreads are already greater than one cent), and displayed size would likely shrink (this latter bit is exactly what happened when prices decimalized). A better alternative would be a tick-size schedule that's a function of price, as is generally done in Japan and Europe.
The "displayed size" would probably shrink, but so what? You'd just need to look at the book to see it.
Decimalization would help even with equities where the natural size > 1c. HFTs who want to get to the top of the book could compete by offering $10.0073 instead of racing to be the fastest at $10.0100. HFTs would compete on price rather than speed.
A more granular tick doesn't just "spread out" the existing liquidity to a bunch of price levels--it meaningfully decreases incentives to post serious size. The spread will end up being marginally tighter, but with thinner books, you still pay more to trade large amounts. The objective function to minimize is transaction costs, not spread.
Since then, we have additional data points from the decimalized US equity markets. See http://www.sec.gov/rules/other/2014/34-72460.pdf for a bibliography. The weight of the evidence points towards thinner books. Incidentally, the SEC is looking to increase the tick size for illiquid small-cap stocks for this very reason.
Because the issue isn't how long it takes for a company to do something. The issue is how fast you can react once information about what they are doing becomes public.
I doubt you will get much of a response from _any_ proprietary trading company. The industry is aware of its public image, and any conversation comes with the perceived risk that you are a journalist looking to write a hit-piece.
Startup costs are a few thousand a month. The firm I worked at has never had more than 3 people and was started by one nerd working out of his flat. You can spend a lot more if you want to do proper latency arb rather than price improvements, but bootstrapping your way to that point is hardly impossible.
I run algo strategies myself - not HFT - and my monthly trading costs are less than my monthly beer costs. Capital invested is perhaps 1-2 years savings for any software engineer (e.g., my best recent trade was dropping $40k on SPY when it was at $190, it's now at $201.66).
Good question. HFT is a spectrum. At the lowest latency (approximately 5 microseconds), the barriers to entry are massive - multi-million dollar startup costs.
For "kind of fast" - 10-100 microseconds - there are a variety of brokerages that can get you started with costs of approximately $1000-10,000 a month. This is rapidly changing though - the exchanges have been continuously increasing the prices of their market data to the point where consuming the entire US equity market proprietary feeds costs around $50,000/mo in licensing.
It's not an especially high barrier if your strategy doesn't require you to build physical infrastructure (most strategies don't require this). You have programmers, hardware, colocation costs, direct exchange connection costs, trading fees, etc. The programmers are by far the most expensive component.
Is that true? Isn't there a high barrier of entry? I was under the impression that large trading firms were building high-speed connections, which is obviously not something a small firm could ever do.