My original interest in physics was in the area of "complex systems". Complex systems are systems that consist of many interacting components and that exhibit emergent properties, i.e., interesting properties that emerge from --- but are different than -- the properties of the individual parts.

A simple example is schooling fish. Fish that link their movements together within a school behave as if they were a single large organism. The properties of the school are unique and extremely important biologically, but they would be very difficult to discover if you only studied the properties of each fish separately.

I believe the economy is a complex system, and I am interested in discovering, and ultimately in explaining, the emergent properties of the economy. It is more of a top-down approach -- from the properties of the system down to the components -- rather than a bottom-up approach that many other economists pursue.

Yes.  Price itself is an emergent phenomenon of financial markets -- it is the result of many people interacting with each other in an uncoordinated, unrehearsed, and unpredictable way.  Despite the complexity involved in these interactions, prices exhibit very interesting and regular properties.

For example: (1) price fluctuations for traded items are almost always statistically random, (2) extreme price fluctuations (such as market crashes) occur universally for traded items and are statistically identical across many different securities, (3) prices respond to trades in a very specific way: they respond more to the first part of a trade than to the last and then they partially revert after the trade completes, and (4) prices of related securities move together in a seemingly coordinated and synchronized way (which I believe is due to HFT activity, even if not by the explicit intention of HFT firms).

No, I don't think there is a simple answer. Some forms of HFT are good and some forms are bad.

In general, those forms that are good tend to connect investors through time and across similar assets, and those that are bad use speed to interject themself between investors who would already find one another if HFT wasn't present.

I think it is extremely difficult to tell the difference between these activities, and that you have to use other evidence -- such as transaction costs, price synchronization, and price accuracy -- to determine whether the good outweighs the bad.

If by costs, you mean investment costs of high-frequency trading firms, then yes, I do think these costs might eventually outweigh the benefits to society.  HFT firms invest heavily on extremely fast telecommunication links and equipment which make markets faster, yet we do not yet know how much benefit each incremental increase in market speed brings to regular investors.  I have, however, started to attack this question in my research.  My colleague, Daniel Fricke, and I recently completed a paper that is a first attempt to analyze the sweet-spot for speed in markets.

I wouldn't say that HFT is only interested in trading liquid securities, rather, I would say that HFT activity is heavily concentrated in securities that require high-speed trading.  In other words, HFT is present where its services are needed and where it can be compensated for providing these services.

For example, imagine a security that is not very volatile, is traded infrequently, and is uncorrelated with every other security in the market.  I can provide liquidity for this security very easily by myself (I could be a market maker), without the need of a computer or high-speed access to the market.  Perhaps not surprisingly, these type of securities do not have much HFT activity.

Now, imagine a security that is extremely volatile, is traded frequently, and is strongly correlated with many other securities.  I simply cannot provide liquidity for this security without the use of (1) a computer to do quick statistical calculations and (2) a high-speed connection to various markets to monitor and react to changing market conditions.  Perhaps not surprisingly, these types of securities tend to have a lot of HFT activity.

To understand the effects of price synchronization, it is useful to draw an analogy to animal groups. Animals that move together in groups -- such as schooling fish -- scan their environment with "many eyes", which allows them to quickly evade threats or find potential food sources.

 

Financial markets are similar. In markets, the state of the economy is monitored by a large number of investors who quickly broadcast any changes to each other and the rest of society via price movements.

By synchronizing prices, HFT allows the "many eyes" of different investors to function as one coherent group. This leads to efficiency gains (just as it does for animal groups) so that prices are more accurate and transaction costs are reduced.

I agree that HFT rarely brings new information to the market by itself. Instead, it very quickly spreads information through the market as it arrives via the orders of investors. It is linking together the information of separate "eyes".

The orders of investors (especially large orders), carry very useful information, and indeed are "real signals". Buy orders tend to arrive when prices are too low and sell orders tend to arrive when prices are too high. This correlation between orders and mispricing is well documented.

This is a good question. No one knows exactly how accurate prices are, but we can use relative measures to determine relative accuracy, and that is what I have done.

Here is a simple way of describing the measurement: Suppose you trade many different times in the market. If the prices of your trades were consistently different than the market price, then you would have reason to believe your trades were priced incorrectly. If transaction prices were accurate and fair, then your trade price should be the same as the market price.

To determine the inaccuracy of transaction prices, I measure the distance between transaction prices and market prices, where market prices are measured one minute after the transaction. I've found that transaction prices are much closer to market prices now than what they were in the past.

Yes, it is possible that something else is causing this increase in pricing accuracy.  However, other academic papers have specifically connected HFT activity (or proxies for HFT activity) to increased market efficiency, and there are currently no alternative explanations.

Technically,  HFT cannot "front run" someone else's order because HFT has no way of seeing an order before it reaches an exchange. They can guess that an order arriving at a certain exchange will be followed by more orders (potentially at other exchanges) but there is no certainty involved.

The uncertainty involved is exactly why it is difficult to determine if this type of HFT trading is good or bad. If more orders in the same direction always arrive and if there are always non-HFT orders resting on the other side of the market that HFT picks off, then yes HFT has needlessly interjected itself between a buyer and seller. It is not technically front running, but it is not good for the market.

However, the above scenario involves many "if's", and in all other cases HFT has provided a service to the market -- it has allowed an investor to transact at a fair price even though a non-HFT counterparty was not present.

As I mentioned before, you really need to look at measures of market efficiency -- such as transaction costs, price synchronization, and price accuracy -- to determine if the good outweighs the bad. The evidence suggests the good does outweigh the bad.

Yes, this is brought up in academic circles as well and is known as the Grossman-Stiglitz paradox. Simply stated, the paradox is the following: If the market shows correct prices, then no one will spend time and money to determine the correct values of assets. But if no one determines asset values, then how can market prices be correct?

The accepted answer to the paradox is that there must be some inaccuracy in the market to entice active portfolio managers to gather useful information about asset prices. These portfolio managers provide a very useful service to the market and are rewarded for their efforts.

I agree with all of this. My point is that when prices are synchronized, information that arrives via the trading of a single portfolio manager quickly propagates throughout the entire market. Because the information of active managers is shared more efficiently, you don't need as many of them to maintain the same overall accuracy of prices in the market.

It is exactly analogous to animal groups that synchronize their behavior -- these groups can expend fewer resources on information gathering, yet still make very accurate decisions.

Yes, I also study extreme price movements in markets. These extreme events tend to be a feature of anything that is frequently traded, whether or not HFT is present.

Unfortunately, I haven't yet studied the relation between extreme price movements and HFT, and I don't know if HFT makes these events more or less frequent.

My work on extreme price movements has not focused on high-speed price fluctuations but more on fluctuations that happen day-to-day or week-to-week.  Market crashes at these time-scales have always occurred.

Although I think HFT can propagate mistakes very quickly in markets -- with potentially disastrous consequences -- I don't think that HFT strongly influences medium to long term volatility, nor do I believe it can create (or stop) crashes like 1929 or 1987.

I agree that controlling a complex system such as the market can be very difficult and that adding new regulation can lead to unintended consequences.

 

However, I think regulation (at least in the US) sometimes contributes to the complexity. For example, the requirement that different US markets cannot display locked prices creates a huge coordination problem that is difficult to solve at ultra high-speeds.

Regulation NMS in the US contains two rules (610 and 611) that, in theory, force exchanges to coordinate together so that investors receive the best price in the market when they submit an order.  (In short the rules do the following: quotes that lock or cross other exchanges are not allowed and marketable orders must be routed to another exchange (or killed) if that other exchange is offering a better price.)

Exchanges, however, are geographically separated from one another so that their coordination cannot be exact.  Sometimes orders that are routed to another exchange do not get filled and would have been better off not being routed.  Sometimes the quotes on other exchanges have disappeared so that locking or crossing would not occur, even if it looks like it would.  Finally, because of maker-taker pricing, many traders want to display a quote that locks the quote of another market (i.e., is at the same price as the other quote), but this is not allowed.

To relieve the frictions caused by these rules, exchanges have created a myriad of order types that have increased the complexity of the market.  HFT firms often use these new order types.

Repealing rule 610 and 611 is not necessarily a straightforward thing, however, as it forces established exchanges to compete with new exchanges when those new exchanges display better prices.  It also forces brokers to do what they should be doing anyway: routing orders to exchanges that display the best price.  In the past, collusion was a real problem in markets, and Regulation NMS has produced markets that although complex, operate very, very efficiently.

Yes, HFT traders cancel many of their orders, but that is not necessarily a bad thing.  In fact, a high cancellation rate is a natural consequence of a healthy market with narrow spreads.

A liquidity provider simply will not quote a narrow spread if they are required to keep their quotes active for too long.  As an extreme example, suppose you were required to post quotes that lasted the entire day.  How wide would your quotes be?  Very wide, close to the typical price change you would observe over an entire day.

In general, I think introducing a minimum resting time for quotes and charging for quote updates is a bad idea.

Yes, the returns to HFT should diminish over time simply due to increased competition. However, HFT firms that provide liquidity will continue to exist in the market (and continue to be a large share of the market) simply because there is a very real need for their services.