Much of the meat of the rule is a bunch of qualitative and quantitative information that banks must collect and hand over to regulators, each piece of which tends to indicate whether a trading desk is more prop-y or flow-y. So if 90% of your trades face customers, that looks like market-making; if 90% of your trades face other dealers, that looks like prop. But there are no bright lines on what is and is not allowed – you just report statistics and hope that the regulators are okay with it.
One important complex of tests involves the distinction between revenues that come from “portfolio profit and loss,” i.e. securities going up or down in value, and “fee” and “spread” income. Fees and spreads are okay. Portfolio is not okay. Or okay only in moderation. Unspecified amounts of moderation.
I continue to have an unhealthy fascination with exactly how the regulators draw those lines. Here’s some of the definition of “Spread Profit and Loss,” from Appendix A, Part IV.B.4 (page 259 of the Fed’s release):
Description: For purposes of this appendix, Spread Profit and Loss is the portion of Portfolio Profit and Loss that generally includes revenue generated by a trading unit from charging higher prices to buyers than the trading unit pays to sellers of comparable instruments over the same period of time (i.e., charging a “spread,” such as the bid-ask spread).
General Calculation Guidance: Spread Profit and Loss generally should be computed using data on the prices at which comparable instruments are either bought or sold by the trading unit, as well as the turnover of these instruments. Spread Profit and Loss should be measured with respect to both the purchase and the sale of any position, and should include both (i) the spreads that are earned by the trading unit to execute transactions (expressed as positive amounts), and (ii) the spreads that are paid by the trading unit to initiate transactions (expressed as negative amounts). Spread Profit and Loss should be computed by calculating the difference between the bid price or the ask price (whichever is paid or received) and the mid-market price. The mid-market price is the average of bid and ask.
For some asset classes in which a trading unit is engaged in market making-related activities, bid-ask or similar spreads are widely disseminated, constantly updated, and readily available, or otherwise reasonably ascertainable. For purposes of calculating the Spread Profit and Loss attributable to a transaction in such asset classes, the trading unit should utilize the prevailing bid-ask or similar spread on the relevant position at the time the purchase or sale is completed.
For other asset classes in which a trading unit is engaged in market making-related activities, bid-ask or similar spreads may not be widely disseminated on a consistent basis or otherwise reasonably ascertainable. . . . In such cases, the trading unit should calculate the Spread Profit and Loss for the relevant purchase or sale of a position in a particular asset class by using whichever of the following three alternatives the banking entity believes more accurately reflects prevailing bid-ask or similar spreads for transactions in that asset class . . . .
And then there are some methods for getting at “typical” bid-ask spreads.
At some level this is very sensible. If you have a 104-104.5 market, get hit, buy a bond at 104, don’t resell immediately, your market drifts up to 106-106.5, and you get lifted at 106.5, your profit is 2.5 points, 2 of which are intuitively attributable to price appreciation and 0.5 of which are attributable to spread. (The rule would treat that 0.5 points as two profits versus mid levels: you made 0.25 spread profit on the purchase and 0.25 on the sale.)
But on another level it’s kind of disturbing, as it only attributes to spread income the “prevailing” spread. In other words, it tells banks that the amount that they can get paid for providing liquidity is the amount that everyone else gets paid for providing liquidity in that particular product.
In liquid markets – listed equities, say – where spreads are small and standardized, this doesn’t do much damage. If everyone’s market is $0.01 wide, then attributing $0.01 (or, technically, half of that, since the "spread" is measured from trade level to mid) to spread is just arithmetic.
But for a high-yield bond that trades once a month, notions like "spread" are pretty loose - banks want to talk sellers down to the lowest price they can get, and then talk buyers up to the highest price they can get. The difference between those is in some sense "spread," and a good trader is one who is good at maximizing it by persuasion, cajoling, strategic use of information, etc. The reason that dealers make markets in illiquid securities is because they have some chance of buying for a lot less than they sell - not because of broad-market price appreciation, but because they can find the cheapest seller and highest-paying buyer.
It would be, let's say, unusual to suggest that Apple should only make as much profit on an iPhone as the average manufacturer makes on the average smart phone. But standardizing spreads does much the same thing for market making activities. Penalizing banks, even in some loose holistic facts-and-circumstances way, for making profits beyond the "prevailing" spread reduces the incentives to make markets - to find the cheapest seller and the highest-paying buyer - in less liquid markets. And if market makers serve a social function, it's to make markets in securities that are less liquid - to find the buyers and sellers who would otherwise be hard to find.