By Kerr Hatrick, Danila Deliya
In the last 18 months, the credit crunch has distorted equity market conditions significantly, but there are some trends that appear to have continued, despite recent shortfalls in liquidity. To provide some context on these trends, we analyze and quantify the mechanics of trading over a significant period of time, across a wide range of different markets. We confine our study to looking at trading patterns of the most liquid stocks1 in the countries studied.
One of the clear advantages of this cross sectional and historical approach is that it allows the identification of outliers. For equity trading, the clear and consistent outlier is still the US, where trading in the most liquid assets is still faster, and smaller, than the busiest non-US assets.
We attempt to show how significant differences in trading environments are, by looking in detail at the month of October in 2008. We compare all venues analyzed and attempt to place them on the evolutionary line travelled by the NYSE. While interesting, we note that such a comparison implicitly assumes that exchanges will travel the same line. At the very least, the regulatory regimes in different regions should make us question this assumption.
Lastly, we look at bid offer spreads and how they have evolved. These spreads are an important part of transaction cost. We look in particular at how they have evolved over ‘Crunch’ period and beyond.
The Evolution of Size and Speed in Global Equity Markets
Over the past ten years, with the growth of program trading, algorithmic trading hubs have moved from novelty to ubiquity. They now manage a sizeable portion of trading activity in major markets. In doing so, they have transformed what was in effect a paper driven process, where traders would calculate what to execute when, into a fully automatic one. Transaction costs have been pushed down to the limits imposed by profitability. What has this process looked like in terms of trade speed, and size? We consider speed first.
We measure speed in terms of typical intertrade duration; that is, the typical time, in seconds, you would expect to wait for one trade to follow another. Results are aggregated over the countries, for the top ten most liquid stocks, per year. We concentrate on continuous, automated trading.
For the US, typical trading duration has compressed steadily. Intertrade times for liquid assets are typically less than 1 second. Indeed, for NASDAQ, intertrade duration for liquid assets is significantly less than this. In terms of the greatest increase in trading speed – and possibly the greatest increase in automation – Europe stands out.
How soon will my exchange be as fast as the NYSE?
Against our better instincts, we measure trading speed on different exchanges, but using the NYSE as a benchmark. In other words – again, for liquid names – if you had to guess where an exchange placed against the NYSE, using just typical trading speed as a measure, which year of the NYSE would be most similar to where the exchange is, currently. Why not NASDAQ? Simply because trading speed there, as measured by intertrade duration, typically makes the other exchanges look unflatteringly slow. For instance, in ’06, the typical intertrade duration on NASDAQ, for our collection of liquid stocks, became faster than all other exchanges are now (including the NYSE). We hasten to add that faster speeds, of course, do not necessarily equate to more efficient markets.
In addition, an increasingly large proportion of NYSE stocks are traded on alternate, non-NYSE venues – so the typical intertrade duration is actually an overestimate. But it seems clear from the data that principal US markets currently outstrip other markets. And in the US itself, one market – NASDAQ – still clearly outstrips the other. The cold hard quantitative facts justify them being regarded, at least currently, as the most evolved, with respect to speed and cost.
How much are trade sizes shrinking, really?
What conclusions can we draw from the data on trade sizes presented in figure 2 and table 3? The years in which trade sizes actually increase – especially in Hong Kong and London – have been the years of the bull market.
Typical size traded is determined by the cost of trading. In the US, the cost of trading on both NASDAQ and the NYSE has been low, relative to their peers in other regions.
Lower costs translate straightforwardly to smaller trade sizes. Hong Kong, on the other hand, stands out as having larger trade sizes. This isn’t just a simple matter of stamp tax, which is levied on a per-notional basis, or the bull market. Trading lots are typically larger in Hong Kong, and push up typical trade sizes.
While there seems to be a general downwards pressure on trade size, the different countries occasionally buck the trend – witness the recent increase in trade size on the Tokyo Stock Exchange, for instance. It should be noted that the data sample, per year, focuses on the ten most liquid stocks traded on the respective exchanges; the further one moves away from liquidity, the greater the uncertainty in the estimation of typical trade size.
Trade Size is Shrinking, but the Queues are lengthening…
So far we have focused on typical intertrade durations, but this is only one measure of the speed of trading. More sophisticated measures also take into account orderbook queue dynamics; in other words, how many times you would expect an order to be placed, or cancelled, on the order book. We measure simply how frequently the bid/ask queues change, over a typical trading day, for our set of exchanges. The order cancellations or submissions we count occur only on the most competitive bid and ask levels. High numbers of changes are a strong indication of significant systematic, or algorithmic trading activity. It is here, in particular, that US trading really distinguishes itself from the rest of the world; specifically, trading on NASDAQ. In Figure 3, we look at a near current picture of number of trades, and the number of bid/ask changes, during October 2008.
Algorithms jostle for precedence in the bid/ask queues, particularly in their passive cycles. During these, they try to execute parts of their original quantity without resorting to a market order. This results in longer natural queue lengths, but does not fully explain the extraordinary number of changes evident on NASDAQ. We attribute at least some of this to other active systematic trading systems, perhaps designed to capture spread.
Ease of Trading over the Credit Crunch
Algorithms which target VWAP typically have a number of hard decisions to make. One of these, is when to transition from passive behaviour to more active behaviour. Optimal passive behaviour ensures good order placement in the bid, or offer queues. In a liquid market where there is some appetite for risk, passive cycles make executions easy and cheap. In less liquid, risk-averse markets, active behaviour is necessary to execute target quantities of shares: the bid-offer spread need to be crossed, and this pushes up the cost of the average execution.
The typical bid-offer spread that must be crossed at different periods of time varies widely. We show how it has varied over the credit crunch period in figure 4. We choose to focus only on this period to minimize the effect of tick size reductions which complicate earlier pictures of spread evolution. Some events in the history of the credit crunch are clearly visible in the typical bid/offer spread of liquid names, trading on the exchanges considered. Bear Stearns hedge funds’ bail-out is clearly visible in the NYSE spread history. Massive credit events like Lehman Brothers’ bankruptcy have a longer lasting effect on spreads.
Throughout the contortions, one trend has held steady: the trend towards faster, and – mostly – smaller and more fractured executions. It has held steady in all markets, through high volume panics, as automation suffuses equity trading more broadly each year.
To end on an optimistic note, if one views bid-offer spread as a risk indicator, then the risk aversion of the market has moved steadily downwards since March of this year. Perhaps new algorithms to deal with volatile spreads will emerge just as the volatility is decreasing. Whatever, it seems certain that algorithms will continue to evolve to the limits imposed by trading costs, and lack of alpha forecasts.