Hacker News Reader

Comment by tptacek | Hacker News Reader

How can a market maker be the fastest without providing liquidity? An HFT market maker's willingness to buy XYZ from Alice at $10/shr to allow her out of her XYZ position is practically the definition of liquidity; its willingness to turn around and sell XYZ at $10.05/shr to Bob is more liquidity still. Alice wants to sell and can do so immediately and at a price she accepts; Bob wants to buy and can do so immediately at a price he accepts. That's liquidity.

The markets do not do that automatically. Before automated trading, Alice could sell, but would probably have had to do so at significantly less than $10/shr, even if the market was indicating $10 was the true price of XYZ. Bob could buy, but would probably have to do so at significantly more than $10.05/shr. The primary reason for that: human beings skimming from Alice and Bob, because there was no effective way to compete.

View on HN

I think we are talking about two different things. Here is the scenario I imagine. Two market makers, Speedy and Big. They both decide to shave a penny off the spread, but Speedy is faster.

The order book now looks like:

SELL BIG - 1000 $600

SELL SPEEDY - 200 $600

-----

Buy SPEEDY - 200 $599

BUY BIG - 1000 $599

A buy and sell order comes in for 200 shares at market, and SPEEDY makes $200. BIG makes nothing. The problem is that BIG is offering more liquidity to the market (1000 shares vs 200), but all the rewards go to SPEEDY. With a pro-rata, BIG would make $166 and SPEEDY would make $33.

Pro-rata rewards volume, FIFO rewards speed. It is debatable which is better.

It looks to me that in either case, 1200 "shares worth of liquidity" are being offered to the market. If 2000 shares of "buy" comes in, it will soak up their whole "sell".

It also looks like "Speed" and "Bigness" are orthogonal issues; you're just irritated that "Speed" got to the book first. But "Speed" is getting compensated for soaking up more risk; because "Speed"'s order fills first, "Big" has more time to react to negative market changes.

Note that in any case, traders are penalized in a variety of other ways for laying large block orders onto the book.

Fine, SPEEDY gets that 200 share trade, now the order book looks like this:

SELL BIG - 1000 $600 ----- Buy SPEEDY - 200 $599 BUY BIG - 1000 $599

Another buy order at $600 comes in and BIG makes the trade.

Now the order book looks like this:

SELL BIG - 800 $600 ----- Buy SPEEDY - 200 $599 BUY BIG - 1000 $599

At this point, SPEEDY wants to get back in so he offers 200 shares @ $600. The Order book now looks like this:

Buy SPEEDY - 200 $600 SELL BIG - 800 $600 ----- Buy SPEEDY - 200 $599 BUY BIG - 1000 $599

So now SPEEDY is sitting behind BIG and has to wait until all 800 shares of his order soaks up the market.

So, where is the problem?

You are right that the interaction is much more complicated to model correctly multi-period than either of us has described. Either way, it is clear that being faster will produce higher returns for a given algo. The question is how much. Given the extreme things HFT guys do, I think it is clear they understand a tiny increase in speed can result in handsome returns. Is this efficient for the exchange?

What if the exchange traded once every 5 seconds in a blind auction, then there would be no real advantage to speed. Each computer would have 4.5-4.95 seconds to parse the auction results and post new orders.

Would this exchange be more or less liquid? (I don't know, but it is interesting to think about.) The barrier to entry to be a market maker would be lower. The market making code would look more like Poker bot code, rather than what HFT does now.

Reply on news.ycombinator.com