## Monday, February 28, 2011

### What is gained from cross-border exchange mergers?

by Ajay Shah.

Cross-border exchange mergers are in the news. See Indian exchanges must go regional and then global and Global mergers and Indian exchanges, by Jayanth Varma, who points us to LSE and TMX merge by Jeff Carter on Points and Figures. Also see Stock exchange mergers: the fight for global dominance in the Telegraph, and Big bourse mergers are back but hold the hyperbole by Benn Steil.

An article in the Economist, Back for more: Has the global exchange industry lost its marbles again?, is skeptical about various stories that are being told about exchange mergers, but holds forth the possibility that there might be cost savings:

Joining forces does not in itself realise revenue gains or alter this decline. But it may make it possible to combine the technology and back-office platforms being used by different exchanges, cutting costs. Efficiency savings are the one element of the last round of consolidation that did arrive as promised.
Cost savings are being emphasised again now. The Deutsche Borse and NYSE-Euronext combination should yield annual savings of  \$412m, the two firms say, equivalent to about a fifth of the combined entity's pre-tax profits, while the LSE-TMX deal should produce savings of about 7%. In this article, I focus on the question: Is there a big opportunity for reducing cost through exchange mergers? ### Getting a sense of the magnitudes An exchange is an IT system that matches orders. The computational complexity of an exchange is all about taking in a lot of orders per second and computing a lot of trades per second. The output of the IT facility is purely measured by the number of orders that were produced. In the public domain, we see the number of trades, and not the number of orders. Hence, the number of trades is the best public domain source of the size of each exchange, from the viewpoint of cost. To illustrate the magnitudes involved, last Friday, BSE got 34.1 million orders and did 1.94 million trades. This is an orders-to-trades ratio of 17.6:1 -- for each trade that BSE produces, they have to have the IT capacity to process 17 orders. The only way to get up to these kinds of values is by having a good deal of algorithmic trading. The revenue per trade is, of course, very different across countries. In India, the average trade size on the equity spot market is \$500 and the tariff for the exchange is hence tiny: NSE or BSE earn Rs.0.65 or \$0.014 per trade. Using the above numbers, BSE's earning Rs.0.04 or \$0.000795 per order on average. These low low tariffs imply that the revenue, profit and valuation of an exchange in India is tiny when compared with what's seen abroad. But on the question of cost, there is direct comparability: it costs as much to produce a billion trades in India as it does anywhere else.

From this perspective, let's look at the biggest factories in the world that produce trades. This is data from the World Federation of Exchanges, for equity trades on the limit order book, in January 2011:

 Rank Exchange '000 trades 1 NYSE Euronext 1,52,922 2 NSE 1,18,200 3 NASDAQ OMX 1,13,753 4 Shanghai SE 1,04,965 5 Korea Exchange 1,00,221 6 Shenzhen SE 76,268 7 BSE 35,157 8 Tokyo SE 27,557 9 Taiwan SE 20,313 10 London SE 19,132

### Saving money through unification of data centres?

I do believe that in this business, there are economies of scale. To build a factory that produces twice the trades costs less than twice the money.

Does this mean that exchange mergers can create value? Not necessarily.

Let's take one plausible merger from the above. The London SE is a small exchange: they did 19.1 million trades in January. The BSE did 35.1 million trades.

Can one save money by producing 55 million trades in a single data centre? Yes.

Will a BSE+LSE merged entity drop down to one data centre? Of course not! The problem is the speed of light. Today, the conversations between securities firms and exchanges are reckoned in milliseconds. And in one millisecond, light only travels 300 km. So even without reckoning for switching overheads (which are huge!) it is not feasible to unify data centres apart from local mergers such as CME and CBOT.

Since light moves at a glacial pace, it is simply not feasible to beam orders from London to a data centre in Bombay. So even if BSE merged with London SE, there would be two data centres. This limits the cost saving. Until we find a way to speed up light, there is going to be no data centre consolidation in this business, other than within small geographical areas (e.g. within Chicago or within New York).

### Saving money on software development?

Okay, let's look further. Could there be cost saving by building one software system and deploying it twice? We'd still spend money to run two data centres, but we'd have only one expense of building software. Could this work?

It's much harder than it sounds. It is not often that one gets to fully transplant an exchange software system in a new location: all too often, the systems have to be significantly different. Regulatory differences, local preferences, history, what users prefer and are used to: all these shape immense diversity in exchange systems. There can actually be diseconomies of scale, with engineering and political problems of handling multiple versions.

Another key problem lies in the sizing of the software system. An exchange system that works for BSE will generally involve a different set of engineering tradeoffs when compared with the LSE setting. So ground-up implementations could be more efficient. By this logic, there may be a useful role for cooperation between similar-sized exchanges (e.g. NSE and Shanghai), but not across divergent sizes which are more than 2x apart.

When decision makers think `a system' can be readily transported across highly diverse order intensities, without regard for the inefficiencies introduced in this process, I think this has something to do with the lack of engineering backgrounds among these decision makers. On a related note, there isn't much of a role for exchange software as a software product, other than in the zone of tiny exchanges where an android phone will suffice for order matching. By the time you get to anything in the top 20 exchanges of the world, an efficient implementation will involve large amounts of ground-up development.

### A skeptical perspective

NYSE merged with Euronext. Did we see cost reductions? A lot was said about cost reduction at the time of the merger, but I haven't particularly seen evidence of this filtering out post-merger.

ASX-SGX: Will they drop down to one data centre? Of course not. Will they unify systems? What will be the cost of system unification? Does it make any sense to unify systems? It helps that both are similar-sized small exchanges, but the institutional settings are highly different.

NSE and NASDAQ produce a similar number of equity spot trades. In the latest year, NSE spends roughly \$150 million a year doing this, while NASDAQ spent \$850 million. (NSE produces derivatives trades also, and the NSE number includes the cost of the clearing corporation, so the cost-per-trade edge at NSE is probably of the order of 10x when compared with NASDAQ). The two exchanges are similar in size in terms of the trades per second. Yet, this is not an easy merger opportunity. There will certainly be no data centre unification. NSE's knowledge can be used to run the NASDAQ data centre more cheaply, but complex organisational dynamics would have to be navigated in achieving the transition, and this could take decades to pull off. It is hard to get management teams that are able to play for such long-term gains.

Also see Are exchange mergers always good? by Mobis Philipose in Mint.

There is one kind of exchange merger which I have become increasingly skeptical about: one in which a parent foists computer systems upon the recipient. I have started worrying that this is a bit of a con, a method to generate revenues from system sales under the garb of partnership or strategic alliances. This is done to some extent by firms that are primarily in the business of selling software and not in the business of running exchanges. Or, to the extent that high-cost exchanges are able to do this, the systems/software revenues are able to mask the deeper problem of a high cost structure.

I have watched the grand global deal-making between exchanges for a long time. In my reckoning, most of it has been a waste of time and money. As one specific example, in my observation in India, some foreign investments into Indian exchanges has been irrelevant, others have directly done damage. None has as yet helped improve product offerings or cost efficiency.

One contract that comes to my mind as one that really worked was Mutual Offset (MOS) between CME and SIMEX, which was done way back in 1984. This was one deal that really mattered and was a good idea. But it was useful in the age before capital account openness - such connections are less important today when capital flows freely anyway. And, remember that it was a mere contract, it involved no complications of ownership and management. So I do think there will be value if the Nifty futures on SGX, CME and NSE are all unified through a mutual offset system: but this does not require anything more complex than signing a contract.

Jayanth Varma says:

It is tragic that at this point of great opportunities and strategic challenges, the energies of Indian exchanges and their regulators are entirely consumed by the debate about whether exchanges should be regulated like public utilities

I disagree. The global exchange M&A story seems to be overrated, apart from the extent to which systems like MOS which can alleviate home bias (and only require contracts). There isn't much to gain there. On the other hand, the problem of sound regulation and supervision of exchanges in India is a GDP-scale issue. Indian experience and evidence does not support a complacent approach that the regulation and supervision will work out.

### Acknowledgements

My thinking on this was improved through conversations with Ravi Apte and Ashish Chauhan.

LaTeX mathematics works. This means that if you want to say $10 you have to say \$10.