HFTs generally consider shorter timescales (both in the feature set and the forward horizon) than human TA would lean on, and make tremendous use of microstructure information rather than outright price movements.

Having said that, “let’s take difference to EMA” is pretty the one-size-fits-all solution, and there are no shortage of firms looking at minute to hour long trades to gobble up anything human TA practitioners would ever try to trade.

> HFTs generally consider shorter timescales than human TA would lean on

This just further emphasizes the point that retail investors relying on TA are having their lunch eaten. HFTs are picking the minute opportunities because that's all that exists. If there were longer-term strategies that were reliably profitable, do you think they simply ignore them?

sub-minute opportunities are absolutely not all that exist, in many ways it’s the opposite. The hft opportunity space is extremely competitive, and much more zero sum than the mid/low frequency space. A relatively small set of players compete for the lions share of anything one might consider high frequency. The set of trades you compete for and the relevant information to generate forecasts is limited / similar across firms, and table-stakes on engineering side are very high.

> If there were long term strategies that were reliably profitable, do you think they would simply ignore them?

Yes? Mid frequency trading (say holding period of 5 minutes to ~1 hour) is a totally different class of quantitative trading, from the features you use to how you build portfolios (if at all) to how you execute on forecasts. At least one top-of-the-game hft firm (that already relied on forecasts instead of speed) got burned on their first attempt to enter mid frequency.

> At least one top-of-the-game hft firm (that already relied on forecasts instead of speed) got burned on their first attempt to enter mid frequency.

Put differently: "A bunch of smart people who have proven to be able to reliably extract profits from microsecond trades tried their hand at exploiting patterns on longer timescales and were unable to. Even though I have less money, information, access to markets, and am just one person, I believe I can succeed where they failed."

I don’t know why you think I’m saying TA works. The only thing I’ve said is that HFTs are by and large not responsible for trading away those inefficiencies

I’ll add that most hft firms are trying to move to longer time horizons. Some with great success, some mixed, some not at all. There’s not much unclaimed pie left in the hft world