Latency, Order Book Staleness, and Market Efficiency

# Latency (market updates) and link to market efficiency


# Latency (market updates) and link to market efficiency












In the book by Lehalle and laruelle - "market microstructure in practice" -

> "The trading activity of HFT updates limit orderbooks at a higher rate than the round trip for any non-colocated observer. The snapshot of the limit orderbooks you take into account to make decisions does not, in most cases, reflect the state of the book when your order actually reaches it... when the orderbook update frequency f is larger than two times the traveling time information (i.e. f > 1/(2τ)), the decision-making process does not use the proper information to decide where and how to route an order. "

However, we don't see this being used in any research paper as a threshold for market efficiency estimates. From the point of view of a market practitioner, there should be lower bound at which exchanges have to segment their market data feeds to a higher number - such that these feeds are usable by the participants.

## Answer by Jim Broiles (score 1)

https://quant.stackexchange.com/a/79304

"The trading activity of HFT updates limit orderbooks at a higher rate than the round trip for any non-collocated observer. The snapshot of the limit orderbooks you take into account to make decisions does not, in most cases, reflect the state of the book when your order actually reaches it."

The quote can be paraphrased as follows: If you are a trader sitting on a desk clicking a mouse while looking at a screen you are no different than a blind person crossing a 10-lane interstate highway. You are about to be killed.

I speculate that the reason you don't see this in other papers on market efficiency is that most of them don't consider latency factors. In fact, most papers regarding various market analysis simply ignore latency because if they modeled latency the results would be abysmal.

Your argument that exchanges should make the data feed usable by all is neglecting the reality that the exchanges make significant profit from collocation and data/order feeds.

In trading as much as anywhere else you get what you pay for. If you want to operate a serious trading business, you must be collocated. You can pick up the phone and be collocated tomorrow. There are dozens of companies that will rent the server and the rack-space. This is available to anyone who is willing and able to pay for it. But being collocated and trading through IBKR (or other such retail broker) is not going to help one bit. Why? I'm going to let you figure that out. It is part of the journey to becoming a professional trader instead of a hobbyist.

## Answer by quantinho (score 0)

https://quant.stackexchange.com/a/77509

This is not exactly answer but comment's length was not enough.

When you do collocation, there are certain conditions in SLA that you could use as benchmark (e.g. max latency of 1millisec for data feed, 10millisec for order etc.), but if your model takes this into account it would be better to estimate it by yourself and get average instead of worst case scenario.

Exchanges/regulators regularly publish papers arguing why their rules/decisions improves market efficiency/fairness, but you can find papers that argue exactly the opposite. That makes it hard for the researcher to decide which side to pick.

In most situations/models latency increases the uncertainty (of whatever you are trying to do), and that is taken into account by some other factor. Almost every backtest engine I have worked with, has an option for fixed or bounded random latency and usually results are the same when you add/subtract 1 millisecond (apparently in real life orderbooks are not as dynamic in 1 millisecond most of the time).