Joe Stiglitz presented a paper on HFT at the Atlanta Fed conference earlier this week that has received a lot of attention. The paper is worth reading, but I actually recommend Felix Salmon’s synopsis, which breaks out the issues nicely.
I agree with Stiglitz in part, and disagree in part. The agreement is that Stiglitz hits many of the themes of my recent posts on HFT, notably that when there is private information, financial markets are unlikely to reach first best outcomes, and that making welfare comparisons is very difficult: I would say nigh-on to impossible, actually. Stiglitz also recognizes that HFT affects the incentives to collect information, which is another theme that I’ve emphasized.
Where I disagree is that Stiglitz (like DeLong) concludes from these insights that HFT is wasteful and should be restricted. This conclusion does not follow at all, and can be traced to some implicit assumptions about the nature of informed trading by non-HFT traders.
Stiglitz says “HFT discourages the acquisition of information which would make the market more informative in a relevant sense.” And by “relevant sense” he means fundamental information about the real economy. He laments that HFT “can be thought of as stealing the information rents that otherwise would have gone to those who had invested in information.” Further, he criticizes that much of what HFT does is merely accelerate the revelation of this information, and this acceleration is so small that it cannot improve any decision on any margin, and hence the resources used by HFT are wasted.
But this implicitly assumes that the information produced by non-HFT traders, the collection of which is reduced by the “stealing of information rents”, is in fact fundamental information that would improve decisions. But as I’ve noted repeatedly, many of the informed traders who HFT firms sniff out are producing information that does not improve any economic decision on any margin. Getting better information about an impending earnings report can be very profitable, but revelation of this information doesn’t improve decision making.
By assuming that non-HFT informed traders are producing information that invariably improves decisions, Stiglitz misunderstands what a great deal of informed trading is about, and thereby ignores a benefit of HFT order anticipation-based trading, and crucially, of HFT quote adjustments that cause markets to run away from big traders and thereby limits their ability to profit on their information.
One way to think about it is that there is cash flow relevant information, and decision relevant information. Pretty much all decision relevant information is cash flow relevant, but not all cash flow information is decision relevant. One major example is what Stiglitz emphasizes: the slight acceleration of revelation of information. But I claim that a lot of the information produced by institutional traders is of exactly this type. Stiglitz (and DeLong) ignore this, which leads them to biased appraisals of the efficiency of HFT.
That is, once one recognizes that some informed trading is rent seeking, and socially wasteful, “stealing of information rents” by HFT can be a feature, not a bug.
Stiglitz also ignores that even if HFT reduces the amount of decision relevant information produced and incorporated into prices, reducing this source of private information still reduces the adverse selection costs incurred by uninformed investors trading for portfolio rebalancing or hedging reasons. This reduction in adverse selection costs tends to improve the allocation of risk. This benefit must be weighed against any cost arising from the reduction in the production of decision relevant information.
In brief, Stiglitz and I agree that HFT reduces the incentive to collect information. Where we differ is that Stiglitz believes this is an unmitigated bad, whereas I strongly believe that this is totally wrong, because Stiglitz’s characterization of informed trading is very unrealistic. My point is that non-HFT informed trading can be parasitic, but Stiglitz does not recognize this or account for it in his analysis.
Stiglitz also complains that HFT liquidity is junk liquidity. In particular, prices move before large orders can be executed.
This is a variant on the criticism that HFT reduces information rents. Moreover, Stiglitz fails to make comparisons between realistic alternatives. The ability to adjust quotes faster reduces adverse selection costs, and allows HFT to quote tighter markets. Restricting HFT in some way will lead to wider spreads and lower quoted depth. Either way, big orders will have a price impact.
Stiglitz also claims that HFT reduces other, better forms of liquidity. Salmon actually explains this point more clearly:
HFT does not improve the important type of liquidity.
If you’re a small retail investor, you have access to more stock market liquidity than ever. Whatever stock you want to buy or sell, you can do so immediately, at the best market price. But that’s not the kind of liquidity which is most valuable, societally speaking. That kind of liquidity is what you see when market makers step in with relatively patient balance sheets, willing to take a position off somebody else’s book and wait until they can find a counterparty to whom they can willingly offset it. Those market makers may or may not have been important in the past, but they’re certainly few and far between today.
HFT also reduces natural liquidity.
Let’s say I do a lot of homework on a stock, and I determine that it’s a good buy at $35 per share. So I put in a large order at $35 per share. If the stock ever drops to that price, I’ll be willing to buy there. I’m providing natural liquidity to the market at the $35 level. In the age of HFT, however, it’s silly to just post a big order and keep it there, since it’s likely that your entire order will be filled — within a blink of an eye, much faster than you can react — if and only if some information comes out which would be likely to change your fair-value calculation. As a result, you only place your order for a tiny fraction of a second yourself. And in turn, the market becomes less liquid.
These points are pretty dubious. The kinds of market makers that HFT displaces (locals on futures exchanges, specialists, day traders) were hardly characterized by “relatively patient balance sheets.” Their holding periods were also quite short. Indeed, one of the filters academics use to identify HFT traders is firms that end the day flat: this exactly what most locals and specialists strove to do. And most traders that “do a lot of homework on a stock” were not doing so to supply liquidity through limit orders that they did not adjust frequently. Those who do a lot of homework are usually liquidity takers, not liquidity suppliers.
In sum, although Stiglitz’s analytical framework and broad conclusions are correct, his specific conclusions about HFT are not. They are not correct primarily because he has a very unrealistic view of the nature of informed trading. Once one recognizes that much informed trading is a form of rent seeking-the point that Hirshleifer made over 40 years ago-most of Stiglitz’s objections to HFT dissolve. Put differently, Stiglitz is right to believe that the financial sector may be too big, in part because there can be excessively strong incentives to collect information and trade on it, but he fails to take this point to its logical conclusion when evaluating HFT.
I do find it rather odd that strongly left-leaning economists like Stiglitz and DeLong who are broadly skeptical of financial markets focus their criticism on one new feature of those markets-HFT-without considering the implications of their broader critiques of the financial sector. At root, their criticism is that much financial market activity is rent seeking. If you believe that, you have to consider how HFT affects these rent seeking activities. Once you do that, it is impossible to sustain the critiques of HFT, because even if there are rent seeking aspects to HFT, it also can reduce other forms of rent seeking.