Acquire data for historical fee structure of the exchanges

[jf434]

Obtain current and historical fee for the different exchanges, and summarize them in the venue_liquidity

[rcyeh]

For the current fee, go to each exchange's web site and look for this information. Familiarize yourself with the structure. Tape A = NYSE. Tape B = Arca and Amex. Tape C = Nasdaq.

For the historical fee, one way to do this would be to call each exchange and ask for this information.

You can say something like: "We are students at Cornell Financial Engineering Manhattan, and our project is sponsored by Peter Gallo at Cantor Fitzgerald. If there are any questions, you can call Peter at 212-294-7797. We are looking for ..."

If you can get statistics on the number of [one-way] shares that were executed in each category on each day, that would also be interesting, because some exchanges (EDGA, EDGX) have a large number of liquidity indicators, and it may not be clear which categories dominate.

[jf434]

Just to clarify, The exchange tends to give various tiers for different customers who pays up more. When we look at the fee structure, which level do you suggest we should focus on more?

[rcyeh]

Good question. I don't know the answer to this, and it's a good question to ask the exchanges when you call them: what fraction of your volume is contributed by customers at each volume tier? (I think the prices are assessed based on monthly volumes, not daily.)

If I were a customer and I knew I traded X shares per month, and X were close to some small multiple of a tier threshold, then this might affect whether I would concentrate all X on one exchange versus another exchange. Maybe there will be a monthly seasonality effect where order flow condenses onto exchanges with volume tier pricing over the course of the month? Actually, I speculate the number of customers facing this optimization problem is probably small --- customers probably know they are in the lowest or the highest tier.

[kz93]

just for example, NYSE has different fee for liquidity taker and liquidity provider, which should we pay more attention to? and there are a bunch other misc fees like co-lo fee, market data fee, etc this is the link for reference https://usequities.nyx.com/sites/usequities.nyx.com/files/nyse_price_list_9_3_13_-_corrected.pdf

[rcyeh]

The short answer is: ignore the misc fees. We're interested in the landscape of per-executed-share fees (both sides).

When a trade occurs, what has happened is: a liquidity taker's order has crossed with a liquidity provider's order. The liquidity taker is assessed the fee or rebate for the shares taken and the liquidity provider is assessed the fee or rebate for the [same] shares provided.

Let's look at this as a game. Suppose there are two traders (Taker and Provider) and two exchanges (E1 and E2) with taker fees/provider rebates:

E1: 0.0030 fee for taking liquidity; -0.0020 rebate for providing liquidity E2: -0.0007 rebate for taking liquidity; 0.0010 fee for providing liquidity

(Note that as long as the rebate + fee is positive for any transaction, the exchange will make money.)

If the stock is available at the same price at both exchanges, then the Taker would prefer to trade at E2, because it will cost 0.0037 per share less to do so. The Provider would prefer to provide liquidity at E1, because for every share transacted, it will cost 0.0030 per share less to do that compared with at E2. With that differential, why would anyone want to provide liquidity at E2? Well, someone who is willing to buy higher or sell lower would provide liquidity at E2. In effect, the rebate and fee structures add another level of granularity to the quote prices.

If the transaction occurs, then perhaps there is some value to the transaction that overcomes the fee/rebate.

So, I can see that on an instant-by-instant scale, selection of the exchange at which to trade would be based on the quote available at every exchange, which I don't show in the data set I've provided you. Those are available in the data from September 2011 to February 2013. For now, I suggest we look at longer-time-scale aggregates instead of the tick-by-tick model; and then we can go to the tick-by-tick model later, after I figure out how to bring the other data set to you.