Below a certain time accuracy, all trading is zero sum. The only reason to be able to trade in milliseconds is to beat someone else. No value is created.
Suppose instead that exchanges buffered all orders for, say, 1 hour. At the end of the hour they perform a stable matching algorithm to pair up bids and asks, and execute those trades simultaneously.
Are there any downsides to this? It seems like it would eliminate all of this zero sum behavior, and at the same time, increase liquidity, particularly for people who aren't on a fiber line one block from the exchange. You could additionally require all material information to be released at the top of the hour, to ensure a full hour of analysis before any trades are executed.
At the end of the hour they perform a stable matching algorithm to pair up bids and asks
"A stable matching algorithm" is the real trick here.
Incidentally, the fundamental problem of zero sum behavior here is caused by the subpenny rule. Speed would be a non-issue if you were allowed to bid $10.031 in order to beat the faster guy who bid $10.03. Right now you can only bid $10.04, which is often too high.
"A stable matching algorithm" is the real trick here.
Easy. Assume all orders are limit orders, because non-limit orders don't need to exist. Perform the Gale-Shapely stable marriage algorithm http://www.cs.cmu.edu/afs/cs.cmu.edu/academic/class/15251-f1... Where the preference ordering of each person is assumed to be the order that maximizes their return. (The difference in the limit order prices.)
The actual situation is much simpler since all players have the same preference ordering, but this is just to show that it's easy to find one stable pairing.
As a percentage of enterprise value, what is wrong with even larger minimal trading increments? There is no "valuation" resolution on a $1B and up company, that hinges on 1/0000th of a percent measurements in the residual value of the equity. Unless people arey marking to theoretical models purely for derivative purposes and trying to maintain a dynamic hedge with increasingly perfect resolution and without cost.
Larger minimal trading increments increase the cost of speculation, thereby reducing the accuracy of prices. What good reason is there for speculators to pay more for liquidity?
The resolution issue on prices is not clear-cut. Liquidity is good, and I'm happy to have it. But infinite resolution (price, timing) trades do not per-se follow. And they ultimately, themselves, have a feedback loop. INstitutionalized front running (changing examples) is not benign liquidity, for example. Dynamic hedging would at least be a legitimate use, but absence its availability should be priced in accordingly in seperate markets.
In this case, it guarantees that no two people would rather trade with each other than with the person they actually traded with. In that sense, the stable pairing is pareto optimal.
I don't care who I trade with. No one does. It's anonymous anyway, unless you trade with certain official market makers. Also, I don't care if my one order for 400 shares@$100 matches with 1 person selling 400 shares@$100 or 400 people selling 1 share@$100.
Trade matching is not the stable marriage problem. Get a clue.
You certainly care what the price on the other side is if you are getting the spread or some fraction of it. That is what determines the preference ordering to run the algorithm.
Sure, that's stable, but it doesn't have a unique solution. Many solutions are stable. How do you pick from the stable solutions in a way that doesn't just recreate the problem?
For example if you say, "a random trader gets to be the first proposer", then the firms with more money just flood it with trader entries, increasing the chance that they get to be first.
>"A stable matching algorithm" is the real trick here.
What about randomizing bids and asks, and matching them on a first-come, first-serve basis?
Let's say you had bids like this 10, 12, 9, 5; and asks like this 11, 12, 10. You randomize the the bids, and sort the asks. Then you do a linear search for the first ask that is <= to that bid, and that's an order. You remove the bid and ask from the lists, and repeat until bids and asks are removed. If there is no ask that is <= the bid, the bid gets popped.
Even every 5 minutes would be great. Or even every 30 seconds. Microsecond latencies don't do much for investing; it's all about trading. Trading (or speculation) even has value so long as it's pulling information from the future to the present. That's the value it provides. If it's pulling the now plus 1ms future to now, I can't see that providing a lot of value.
With thousands of hedge fund PhDs all trying their damndest to exploit the market, adding delay will change the game but not eliminate it.
Only by when the market is efficient (in the technical economic sense) is it non-exploitable by smart guys. Efficient markets require all players to be rational, so not in this world.
Precisely. Information asymmetry is what allows the exchanges to be profitable. It is a continuous wealth transfer along the information gradient, with the most knowledgeable investors capturing most of the wealth from the least knowledgeable.
As long as this asymmetry exists, the vast majority of trades will constitute a zero sum wealth transfer.
Just because perfect efficiency is not attainable, that doesn't mean we shouldn't strive to get as close to perfect efficiency as we can.
There are issues with regulating markets (serious ones given the imbalance of resources between regulator and banks you point out), but it is not impossible.
There are real needs (hedging, especially if your portfolio has options) to know instantly what prices your actions get executed at.
There are problems with the low latency trading, and some unscrupulous behavior, but it's not entirely bad. In general the low latency guys do provide liquidity, and tighten bid-offer spreads.
The problem with low latency trading is it adds liquidity when you don't need it aka when things are stable and removes it when you do aka the market is crashing. Also, from an information theory perspective high frequency trading can't aid price discovery as the algorithms don't directly act on outside information and there assumptions are hard coded.
Market makers don’t directly participate directly in the price discovery process. A high frequency trader, or any sort of market maker, has no opinion or information on whether Apple is a valuable company.
Further buying or selling within the spread does not automatically make money it's as there is risk which that article mostly ignores.
Keep reading. I (it's my blog) explain how market makers reduce the transaction costs for speculators, allowing them to more efficiently discover prices.
Similarly, network engineers don't gather news, but they still help the news get out.
It assumes that they need to trade right now and sell right now which is a false assumption. As to what network engenders do, if you want to really stretch thinks Starbucks barista's aid price discovery by keeping people awake but that's a meaningless definition.
PS: There was a while that I accepted the idea, but consider what happens if bill gates wanted to sell 10 billion in Microsoft stock HFT will be willing to pick up quite a bit, but vary seen there will try and get rid of the stock and the net effect is they don't hold the stock for long enough to matter. Because, really what they do is time shifting not price discovery. Further the reason it works is buy and sell orders without attached prices which makes selling 1/10th of a second sooner cost you money.
Anyone who doesn't need to trade right now is free to place ALO orders, not cross the spread, and collect a few fractions of a cent in rebates. The fact that most traders choose not to get paid to do this should suggest to you that maybe you are incorrect about the need to trade right now.
I have no idea what you are talking about with "buy and sell orders without attached prices" (market orders?). You are correct that HFTs probably do not provide enough liquidity for Bill Gates to immediately liquidate his entire portfolio. So what?
If all trades are executing at the same time, what do you gain by knowing prices at finer intervals than that? The things you are hedging against would operate the same way.
Currently you can implement "Enter this position in an amount proportional to the portion of that order/those orders that get filled/do not get filled" yourself. You send the first order, you receive the response in a small amount of time, and then you send the order for the hedge. Alternately, you send the order and a hedge of a guessed size concurrently, wait for the order's response, and then adjust the hedge. Five minute or half-hour round-trip times would not be acceptable regardless, so a proposal that seeks to be taken seriously should address this issue.
Some thoughts on addressing the issue: implementing this sort of thing entirely on the server side would be a colossal engineering effort, and probably involve adding new "fill conditioned on state of other order (on a different exchange in a different country)" order types. A naive implementation of these conditional orders as described permits causal cycles where taking the shares causes one to take the shares, and not taking them causes one not to, and then one wonders whether the exchanges (the many worldwide exchanges that are collaborating on this algorithm) should have the order take the shares or not! A similar causal cycle may occur in which taking some shares causes one not to take the shares and vice versa. Regardless of what resolution is chosen for the "paradoxical" case, it will undoubtedly be a significant undertaking for both mathematicians and bakers to discern the correct combination of order types to maximize EV or to minimize risk, and to choose between maximizing EV and minimizing risk each time a position involving more than one instrument is traded.
Alternately, we could allow each exchange to make its own rules. They might end up with rules that are actually possible for human engineers to implement, like "price-time priority".
I meant in particular, it is not okay to have a one hour round trip time between the legs of a single position.
If I want to enter a large hedged position using the same "send some orders, wait for the fills, send more orders to tweak one side to be the appropriate size" procedure, I want to minimize the potential for price movement between the first and second sets of fills. If one hour's worth of information is incorporated into the securities' prices between the two sets of fills, that is a really huge amount of exposure compared to what I would expect today. I would have to be believe that the EV of the position is quite large before accepting one hour's unhedged exposure to get in on it. I could try to get rid of the round trip by using market orders everywhere, but this doesn't actually sound less risky. My belief and knowledge of the trades at some particular time would have to justify entering the position at some unknown pair of prices an hour later.
The great many participants who have not-incredibly-strong beliefs about the value of some hedged position will just not trade. The absence of their trades is a loss of information to the market; the market is less efficient. We probably share the belief that a more efficient market is good and a less efficient market is not so good :)
Enough with the hedged positions though!
Normally for this type of proposal I can just answer "If we do this then spreads will get bigger and that's bad," but here I have trouble doing that, because I'm not even sure if the notion of a spread or an inside price would make much sense once everything is so illiquid. Rather than having a view of the market that is a bunch of orders I can interact with at will, I would instead have a record of all the trades that happened in the last timestep and which orders were intact at the end of the last timestep. This is probably not sufficient information for me to learn the best prices at which I can be certain to buy or sell.
>If one hour's worth of information is incorporated into the securities' prices between the two sets of fills, that is a really huge amount of exposure compared to what I would expect today.
I don't think it's accurate to call what's going on today "an hour's worth of information." Not that much happens in an hour, per stock. It's more akin to an hour's worth of random walk. This is harder to argue, but I think this scheme would cause most of that random walk to disappear. Certainly whatever component of it is second order would disappear, (that is, based on signaling from the price movement itself.)
>I'm not even sure if the notion of a spread or an inside price would make much sense once everything is so illiquid.
If you think of liquidity in terms of "the chances that I can sell without having a large effect on the price of the stock" then I think this only increases liquidity. You're no longer counting on the availability of moment to moment orders on the other side of the deal. You're essentially building a smoothing function into the market.
There would be an incentive to wait for the very last moment to submit your order to such a system to maximize information for that period. However, this would be rarely useful for traders, unless there was some other automated thing, like a secondary market, that determined successful bids and asks.
NASDAQ and NYSE offer intraday crosses in addition to the continuous cross.
Having only discreet crosses would not remove the value in lower latency for trading economic releases. There have been a variety of services (like Pipeline) that have services to facilitate large block trades. I fail to see how this would increase liquidity.
Suppose instead that exchanges buffered all orders for, say, 1 hour. At the end of the hour they perform a stable matching algorithm to pair up bids and asks, and execute those trades simultaneously.
Are there any downsides to this? It seems like it would eliminate all of this zero sum behavior, and at the same time, increase liquidity, particularly for people who aren't on a fiber line one block from the exchange. You could additionally require all material information to be released at the top of the hour, to ensure a full hour of analysis before any trades are executed.