Two previous issues of the Artha delved on concerns relating to high frequency and algorithmic trading. In the article on algorithmic trading (Vol 1 Issue 7) it was mentioned ‘whether algo trading is good for the market is a question yet to be answered’. Later in another related article on high frequency traders (Vol 2 Issue 5) it was suggested ‘capital market regulators in India should not be too much bothered about the abuse of HFTs at this stage since the empirical evidences so far on the role of HFTs are mixed’. Algorithmic trading refers to use of computers and programs to generate and execute large orders in markets with electronic access. Such orders come from institutional investors, hedge funds and trading desks of brokers. Rather than maximizing profits, one of the important objectives of algo traders is to minimise trading costs and market risk. There are two known methods of algo trading- high frequency trading (HFT) and quantitative trading (QT). HFT involves real time orders in milliseconds whereby a trader places and quickly cancels small orders to find out at what price a trade can take place. HFT profits are largely driven by volume. QT is a longer term trading strategy where the algorithms analyze the trends and predictable patterns in the market and trade upon machine-derived forecast. National Stock Exchange (NSE) introduced co-location services (a paid facility) for the traders in January 2010. This facility allows market participants to rent servers located within the NSE’s premises. Co-location refers to bourses allowing members to set up automated trading systems on their premises to reduce latency i.e., the time required for orders to flow between the exchange and the broker’s trading system.1 NSE, on an average, saw 21.67% of its turnover from colocation servers during the first three months of 2015. In many advanced countries, an order is fragmented not only into smaller lots for execution but also is traded through multiple exchanges. In India, on the other hand, top two exchanges BSE (Bombay Stock Exchange) and NSE control almost the entire equity market. NSE, for example, accounts for 80% of equity spot trading and almost 100% of equity derivatives trading.2 Therefore, it becomes relatively easier to study the effect of algorithmic trading on the quality of the market. A central question to the debate of algorithmic trading (AT) is whether AT is beneficial for the market in particular and society in general. A related question could be whether introduction of AT has improved market liquidity particularly during periods of crisis. Liquidity is best represented by the depth (number of shares available) at the top of the book (best bid and ask) as well as at the prices beyond the best quotes. High liquidity is very important to retail investors, especially when market condition is stressed. It is generally believed that algo traders supply liquidity when bid-ask quotes are wider. This is to take care of adverse selection cost- the cost associated with trading against an informed counterparty. Bid-ask spread gets large in case of high market volatility. Therefore, one can naturally expect algo traders supplying liquidity in times of high market volatility. However, evidence in this regard is mixed. While a study3 in the Hong Kong market shows algo traders supplying liquidity in times of short-term market volatility, we have done a pilot study in India and found when volatility is extremely high, rather than supplying limit orders, algo traders become consumers of liquidity. AT and Market Microstructure The message traffic in Indian stock market has increased manifolds since the introduction of co-location facilities in the stock exchanges. Message traffic includes electronic order submissions, cancellations and trade reports. In many markets, it is almost impossible to identify whether a trade is generated by a computer algorithm. Researchers, therefore, use proxies to identify trade. For example, Handershott et al,use the rate of electronic message traffic as a proxy for the amount of AT taking place.
We have obtained a dataset from NSE that has separate flags to directly identify algo trades- a feature not available in similar database of most of the stock exchanges. Average daily messages in NSE is in millions- from 76 million in 2011 it has grown to 171 million in 2013 (Table 1). Messages posted by algo traders’ account for about 95% of the total postings. A large number of orders submitted by algo traders are either subsequently revised or cancelled. The NSE dataset also distinguishes algo trading into proprietary algo (propalgo) and agency algo. Message to trade ratio for propalgo is more than ten-times that of agency algo indicating that prop ATs dominate the high frequency trading market. The order-totrade ratio for propoalgo is also similarly high. SEBI, through a circular in 2012, imposed a fee to be paid by algo traders for high order-to-trade ratio. SEBI is of the opinion that large unexecuted orders create unnecessary pressure on a trading system and also hampers price discovery. SEBI has advised stock exchanges to monitor orders from trading algorithms in order to arrest or identify any market manipulation by such traders. Order-to-trade ratio for non-algo traders is reasonable. AT and Flash Crash A flash crash on the National Stock Exchange (NSE) on October 5, 2012 around 10AM due to erroneous trades by a dealer in 59 frontline stocks pulled the NSE-50 (Nifty) index down 15.5% in 8 seconds5 . Though NSE clarified that the abnormal orders were ‘non-algo’ in nature, skeptics raised the issue of the exchange not able to detect such order on time and to allow the trade to happen. Trading was subsequently stopped and trading resumed at NSE at 1005 hrs and NSE behaved normally thereafter on that day. We looked at the behavior of randomly selected five stocks on that day (Figure 1). All the stock saw sharp fall around 10AM on October 5, 2012.
How did the algo traders behave on that day? Were they providing liquidity around 10AM on that day or cancelling orders? Figure 2 clearly shows that message-to-trade ratio had sharply dropped around the crash time. Reliance, for example, had a message-to-trade ratio of more than 4000 around 9.37AM and when trading resumed at 10.05 AM the ratio was around 14. About 90% of orders (of Reliance) were cancelled by the prop algo traders during the first fifteen minutes of market opening and then it suddenly dropped to almost zero at 9.51 when trading was stopped by the exchange. This shows that algo traders had no clue about the crash and were behaving normally immediately before the flash crash and once the trading resumed such traders again got back to their normal behavior.
Figure 4 indicates proportion of prop algo trading of total traded volume. It is observed that AT volume had shot up around 9.51AM when the flash crash was observed. Whereas the non-algo cancellation (Figure 3)peaked around the crash time. This implies that while the non-algo traders panicked during crash, algo-traders actually traded at a disproportionately higher level.
Note: Reproduced from Artha, May 2015
Article By: Ashok Banerjee and Samarpan Nawn
Professor, Finance and control andFP Student (Finance & Control), IIM Calcutta
On one side, we have the world’s youngest workforce and rich demographic dividend and on the other side we have people with higher education degrees applying for Government jobs requiring much less
Khyati Sharma, a student of PGPEX, has got her fiction cum poetry book shelved at Amazon, Flipkart, Googleplay, Bookadda, Ingraam, dailyhunt, infibeam, Kobo, and Scribbed. The book uniquely chronicles
“A year of being in Joka can be as eventful as you can imagine and a huge learning experience! Choices, managing the highs and lows of a year in a Bschool can teach you lessons for life…” *