From the article and other sources I've seen before, it seems that algorithmic trading is not necessarily high frequency, but that high frequency trading is necessarily algorithmic.
In which case is this not a rather thin hair to split?
Not really. HFT is a subset of algorithmic trading, where many small orders are placed to take advantage of intraday (or intraminute, or even intrasecond) shifts in the spread. A large strategic order executed through an algorithm is a different creature altogether.
The big problem when you place a huge buy or sell order is that it shifts the price in a direction you don't want it to go. For a large sell order, the price goes down, as the market becomes skittish about the security. For a large buy order, the price goes up, as the market becomes bullish and arbitrageurs quickly buy up securities to resell to you. So, many traders use algorithms to hide their trades. The purpose of these algorithms is to avoid volatility, so they shouldn't be dangerous to the market, as long as they're designed correctly.
(Although, to be fair, algorithms may automatically stop trading when the market becomes too volatile, which contributes to flash crashes by reducing liquidity.)
"so they shouldn't be dangerous to the market, as long as they're designed correctly". Even if we accept the author's position, why would we also make the assumption that this software is designed properly and bug free?
I guess it depends on the stakes, and how well the traders understand (the risks of) software development to pay for the quality, testing, and maintenance.
"This isn't just some human lives we're playing with, this is serious! It could cost us billions!"
HFT is about making money by supplying liquidity or making money by exploiting the portions of the system which are designed to supply liquidity. You don't want to end up with a position. You want to net out the day with no position in any stocks. HFT is done by an HFT fund which is usually trading off credit extended by an agressive hedge fund who they split the profits with.
Algos are about getting a trade done at the least cost. You want to buy 100M shares for fund XYZ or sell 10M shares for fund ABC. Algos are run by brokers on behalf of their buy-side clients. A simple algo is something like TWAP, the time weighted average price, which attempts to buy a large position over the course of the day gradually. The purpose of this algo is to avoid moving the market. This is suitable for the "build up a large position because I'm bullish on this industry" type of trade.
Another algo might be IQx, a package offered by CovergEx which tries to quickly execute the trade by sending it out to multiple dark pools and the major exchanges at gradually worse prices until the whole order fills. (Or at least that's my reading of the marketing material.)
Example: there are algorithmic strategies designed to make large trades.
Let's say an institution wants to buy 10K shares of a thinly traded stock (lets say, 100K shares per day). Placing a single order for 10K shares will send a signal to the market that someone wants to buy, and other traders will see it.
There are special algorithms designed to purchase the specified number of shares without making too much of a market impact. For example, in this case, the strategy would send 100 share orders rather than presenting the full 10K interest at once.
for the googlers: there are all kinds of technical terms like implementation shortfall to give more info
In which case is this not a rather thin hair to split?