Interesting item today by a sociologist on The Conversation about the danger of high-speed trades in stock markets.
It contrasts with other reports of studies that conclude;
Economic research thus far provides no direct evidence that high frequency computer based trading has increased volatility.
Of course, the absence of evidence that high-speed trading increases volatility does not mean that it high-speed trading doesn't increase volatility. But the bigger question isn't just what happens to volatility but how markets are affected.
The particular issue is how a trader, automated or otherwise, values a stock. Like any asset there are two sources of future value for the stock. The first is the income stream from it represented by the value of dividends and an appropriate (see note) discounting formula. The second is the likely future price realised when the asset is sold.
The difficulty in real world markets is how much weight should be given to each. More specifically what is the right response for a trader to make as asset prices move without any corresponding change in expectation of future earnings? If prices are increasing do you buy (on the expectation of future increases) or sell (as the price is now higher than the value of future earnings).
Accountants don't help much because they have accounting rules about assets being "marked to market" - which was partly the issue in the GFC as finance houses reported massive profits by revaluing market assets despite the fundamentals not being there.
The second issue is the extent of correlation between asset prices. The GFC in part occurred because of an assumption that not all house prices would decline at once - but they did. That invalidated the risk modelling that underpinned the derivatives.
The same is true of stockmarkets. Fear of a recession will drive down all stock prices because the fundamentals change - recession equals lower profits. But that small shift based on fundamentals then feeds into the future price expectation.
My sense is that speedier automated trading doesn't increase volatility as such. But standard market models tend to ignore the kinds of real world distributions identified by Talebi in Black Swan or Mandelbrot in the Mis-behavior of Markets.
The other unanswered question is whether once markets have sunk into a hole high-speed automated trading makes it harder to climb out of them.
These are issues that I think are better resolved not by statistical analysis of market behaviour but by agent-based modelling of markets. The latter would provide the means to experiment with market rules designed to manage pricing better. An example of a possible rule is to limit the use of "mark to market" in accounting - that assets cannot be re-valued up faster than a rate of double CPI purely on the basis of "mark to market", and that equally they must be re-valued down strictly according to "mark to market." I don't know how well that rule would deal with the issue. That's why simulation would be desirable.
Note: Discounting is used by economists as if it is easy. However, the first thing we know is that real traders apply "hyperbolic" discounting - they overweight short term payments. The second is that the appropriate "discount rate" is really the market price of another asset - Government bonds - plus other unknown risk factors. In other words "discounting" sounds like a scientifically sound procedural method but it is as much gueework (or estimation) as science.
Novae Meridianae Demetae Dexter delenda est