By Mihai Bistriteanu
Mihai Bistriteanu of Citigroup Global Markets Japan focuses on the trading pattern changes post-arrowhead implementation as well as its many opportunities and challenges.
Without a doubt, the Tokyo Stock Exchange’s (TSE) implementation of “arrowhead” this January 2010 was one of the most important events in the history of Japan’s equity trading, with a lot of information and articles being circulated on the structural changes following the new system’s implementation. It is indeed amazing how arrowhead’s implementation changed the very core of Japan’s trading landscape making a huge impact on latency, market volume, trade size, price, as well as tick size dynamics.
Low Latency
Primary exchange speed was along awaited feature in Japan. The one digit millisecond turnaround on a non-collocated infrastructure, promised by the Tokyo Stock Exchange (TSE), is now finally happening in Japan, with a few interesting trends starting to develop as a result of the higher speed:
- Strategies requiring low latency infrastructure can now be easily implemented across Japanese stocks.
- There is an ease of integration of TSE flow within SOR (Smart Order Routing) systems and crossing engines.
One of the key drivers of arrowhead implementation is the aggressive growth of competition in Japan. The PTSs (Proprietary Trading Systems) and broker dark pools had the speed, and in some cases, price as competitive advantages. The liquidity available outside the primary exchange is still small compared to the ratios seen in Europe and the US.
A large percentage of the sophisticated buy-side investors refrained from using SOR technologies to avoid missing liquidity in the primary exchange because of the overall latency.
A standard matching system, with queue jumping functionality when posting liquidity, usually places multiple legs of the same order in multiple venues. When an order is matched in the proprietary crossing engine, the system sends cancellation requests to the other venues, and only when the acknowledgement is received is the cross executed.
The bottleneck for this technology used to be the speed of receiving the acknowledgement from the primary exchange, when the cross was found. Similarly SOR technologies send IOC (Immediate or Cancel) orders to multiple venues including primary exchanges. When one of the venues is slow, it will delay the entire system, therefore missing rapid price changes.
The fact that the TSE infrastructure speed is now in-line with its competitors will have a positive effect in widening the use of liquidity aggregation tools. This also may eventually result in an increase in liquidity on the alternative liquidity pools. The new broker pools and PTSs have to be competitive on all parameters (speed, price and liquidity) to enter and be successful in the Japanese market now.
Market Volume and Median Trade Size
Following the implementation, the volume growth started gradually. At this very moment, as I am writing this article, the rate of growth is still very positive. We expect this to reach saturation over the next couple of months. The main drivers behind the growth are the new strategies that take advantage of the high speed environment, as well as the general increase in activity at the beginning of the year and the market recovery.
The decrease of the median trade size is the natural consequence of the higher percentage of electronic trading in the market. Though, this has been happening for several years now, what is different this year, is the steeper adoption. This is caused by the chain reaction effect (Figure 1):
- New flow generated by high frequency models with tiny order sizes are bringing down the average trade size in the market;
- Some of the algorithmic models are sizing their splits in relation to the average trade size in the market, consequently generating smaller splits
- Human traders having difficulty trading manually are sending more flow to the automated systems, further bringing down the average order size. The transaction data is increasing dramatically and greater investment will be required in scalable, front to back office technology. (Based on a rough estimate by the Exchange about 3.5 times higher data volume was experienced in January 2010 versus December 2009.)
Tick and Price Dynamics
Volume shocks now have a larger impact, resulting in steeper price spikes. Within a shorter period, the stock price can reach ‘limit down’ and eventually bounce back, if the swing was caused by a larger impact order. Exhibit 2 shows an example of a stock price fluctuation on January 7. The effect is amplified now, as the matching speed changed from once every 3 seconds to immediate execution. In addition, the speed of ‘participation strategies’ is higher and the “follow the shock” catch-up effect more important.
Once a large order moves the stock price, the immediate effect is a sudden widening of the bid/ask spread. This could be used as a filter to avoid trading immediately after a volume shock. However, in Japan, due to the long queues for some stocks, many strategies are placing orders to get the maximum priority without crossing the spread. It is only required to have two different strategies with similar trading pattern on the same stock to narrow the spread immediately.
Let’s consider a fictitious example as shown in Figure 3. We have a stock trading at 100 bid/ 101 ask; a sudden large volume comes into the market and moves the stock to 96 (-4%). We assume that the offer at 101 is constituted of 2 different strategies with similar behavior and let’s call it for the sake of simplification ‘Be the Near Touch’. Both orders’ (100 shares each) target is to always get the maximum priority, and become the ‘near touch’ when possible.
Following the spread widening, the second order in the queue will immediately amend down, to become the offer with the highest priority (at 98 for example). The order left behind will follow the amending at a lower price to achieve the same target. In a matter of milliseconds, the spread is artificially brought down to one tick at the 95/96 level, making it impossible for the filters based on the sudden spread widening, to work properly. Eventually the price will move back at 100/101 level, however this is a significant volume traded 4% below the expected price.
To avoid this, traders have started to use shorter term VWAP, instead of IS (Implementation Shortfall) or POV (Participation of Volume) strategies to steer away from over-participating at non-favorable prices following volume shocks. As a result, the shift from VWAP/TWAP strategies to IS and liquidity seeking strategies that has been happening over the last few years in Japan has reversed in January ‘10. More sophisticated, anti gaming and “follow the shock” algo trading filtering features will shortly become available to help cope with these problems. Once implemented across the main algo providers, the shift to IS and liquidity seeking strategies will continue.
Spreads
The most positive change is on the trading cost, as there is a direct relationship between execution cost and bid offer spread. A ‘medium performance’ VWAP engine has a slippage of about 20% of the spread. The reduction of 25% of the average daily spread of the Nikkei 225 will have an overall and directly proportional effect on the trading costs. The dramatic change is for stocks within 2,000 to 3,000 yen range, which are now five times cheaper to trade.
Conclusions
We have seen a very rapid evolution to market-microstructure in response to the system changes. Spreads are down, median trade sizes are down, transaction numbers and market data volumes are up. Most significantly, the tick inside touch, a defining characteristic of the old microstructure, has been replaced with a normalized, smaller touch with the bulk of the volume a tick or two behind the market. Participants who have invested in systems capable of effectively trading in this more efficient environment will have a significant advantage over those who have simply connected old systems to the new market.
