by Nidhi Aggarwal and Chirag Anand
SEBI recently released its discussion paper on algorithmic trading. The paper proposes several measures to address various concerns that have been expressed about the rise of new technology in the field of financial markets. One of the candidate interventions that SEBI has proposed is the imposition of ‘minimum resting time for orders’. SEBI proposes imposing a resting time of 500 milliseconds (ms) during which an order will not be allowed to be amended or cancelled. In this article, we bring evidence to bear on this one candidate intervention.
The rationale for the proposed measure is to curb ‘fleeting orders‘ or orders that appear and disappear within a very short period of time. SEBI’s proposed regulatory intervention, that there should be a minimum resting time of 500 ms, may suggest that orders modified/cancelled in less than 500 ms are considered by SEBI to be fleeting orders, though the discussion paper does not say this explicitly.
A central objective of the regulation of financial markets is to block market abuse. How can fleeting orders be connected with market abuse? Orders without a clear intent to trade may falsify perceived liquidity and price in the marketplace. Through this, placing fleeting orders could be a tool for misleading other traders. In the market abuse literature, there is a concept known as “order spoofing”. This involves placing a visible order in the opposite direction of the trade that is genuinely desired. For example, a seller might post a small buy order priced above the current bid, in the hope of convincing other buyers to match or outbid this. If that occurs, the trader can then sell into this (higher) price.
Fleeting orders can also contribute to “quote stuffing”, which can affect the ability of other traders to send their orders to the exchange by essentially flooding the systems. This is tantamount to the strategem in the field of computer security that is called `denial of service attacks‘.
There are, however, legitimate and important reasons for rational persons to place fleeting orders. A trader may cancel and resubmit a limit order when the market moves away from the original limit price. This will especially occur during volatile times when information arrival in the market is high. Many trading strategies look at the touch — at the bid and the ask price — and not just at the last traded price.
A perfectly legitimate trading strategy runs as follows:
Watch the bid and the ask price, continuously compute (bid+ask)/2 which is the reference price, always have a limit order to sell at 0.1% above this reference price.
A person may use such an algorithm to sell a large block of shares while hoping to get a good execution price. This algorithm would dance continuously, refreshing the limit order every time (bid+ask)/2 changes, which is much more often than a change in the last traded price. That is, this trading strategy would undertake more revisions per unit time when compared with the number of trades per unit time.
Every now and then, a trader might just switch a limit order to a market order to get immediate execution (Hasbrouck and Saar, 2009). This would look like a fleeting order as the trader changed his mind and scrapped a limit order after a very short time.
Before designing an intervention, SEBI needs to examine the data to look at the fraction of and nature of fleeting orders in the market. The discussion paper has no evidence about the existence of fleeting orders, or evidence that there are problems in what is going on at Indian exchanges. Without a clear demonstration that the issue exists, the coercive power of the State should not be used. If we go down the path of using State coercion without the foundations of hard evidence, then there is a high chance that State power will merely reflect competing political pressures where various factions try to use State power as a tool for furthering their business objectives.
In this article, we analyse the questions surrounding fleeting orders using data for orders and trades from the National Stock Exchange of India.
The database and computational challenges of such work are immense. Hence, we use two months of data for the analysis.
Ideally, this work should have been done using data for June and July 2016, but our computational infrastructure broke down in January 2014, and we were forced to make do with the most recent available complete months, which were November and December 2013. The intensity of algorithmic trading in November and December 2013 is the same as that which has prevailed in the following months. Hence, we are on sound grounds when we analyse that data.
There were 6.5 billion records of data in these two months, which were studied for the purpose of this article.
A large fraction of orders on NSE are cancelled. In an analysis that we did in July 2015, where we studied the same months of November and December 2013, we found that 56.97% of new orders that entered the spot market, 94.11% of the orders on the single stock futures (SSF) market, 88.55% of orders on the single stock options (SSO) market, 82.58% of the orders on the Nifty futures market, and 87.51% of the orders on the Nifty Options market were cancelled.
Order cancellation is clearly a valuable tool for most traders on electronic markets. This is seen internationally also. For example, Hasbrouck and Saar (2002) find that 93% of limit orders are cancelled on INET. This is true for other exchanges including NYSE, the Australian Securities Exchange and so on.
In a deep sense, algorithmic trading is merely trading by other means. Using data from NASDAQ, Subrahmanyam and Zheng (2015) document that cancellation ratios of high frequency traders are similar to that of the non-high frequency traders.
A large percentage of cancellations does not imply the existence of fleeting orders. A fleeting order involves placing a limit order inside the touch (i.e. between the bid and the ask) and then quickly cancelling it (Fong and Liu, 2010). There are three steps in identifying fleeting orders: We must count (a) Cancelled orders, (b) Which were cancelled quickly and (c) Which were near the touch. We will now do these calculations for the NSE spot and SSF markets.
Duration of cancelled orders
We analyse the securities which were traded on the derivatives market in 2013. These were the top 150 firms. We group these securities by market capitalisation. The securities with the highest market capitalisation are in Q1, and the securities with the lowest market capitalisation are in Q4. For each quartile, we measure the fraction of orders which were cancelled. We go on to measure the fraction of these order cancellations which took place in under 1 second. This is a conservative value when compared with SEBI’s proposal of 0.5 seconds. If SEBI’s proposed threshold of 0.5s were used, the fraction of orders seen would be lower.
All values as % of total unique orders entered
|Panel A||Orders cancelled|
|Market Cap Quartiles||Spot||SSF|
|Panel B||Orders cancelled in less than 1 second|
Panel A of Table 1 shows the share of cancelled orders in all unique orders, while Panel B shows the share of orders cancelled within one second of arrival in all unique orders. We see that in comparison to the SSF market, the spot market experiences a lower percentage of orders cancellations within one second of arrival. In addition, we see that the percentage of order cancellations within one second is higher for large market capitalisation stocks. This is consistent with the fact that the biggest firms are the subject of the most intensive scrutiny by the financial markets.
The biggest value in Panel B is 70.06%: A full 70.06% of the SSF orders for top quartile stocks are cancelled within 1s. The smallest value is 12.60%: Just 12.60% of the spot market orders for bottom quartile stocks are cancelled within 1s. We should not that this is the bottom quartile within the top 150 stocks on NSE, i.e. it is the stocks from rank 113 to 150.
We now turn to measuring the extent to which these fast cancelled orders could be termed fleeting orders. We only focus on the orders cancelled within one second of their arrival.
Position of ‘fast’ cancelled orders before exit
The table below shows the position, in the limit order book, of fast cancelled orders before they were cancelled. ‘At best’ indicates that the order was at the best prices in the book, (1,3] indicates that the order was placed at depth two or three in the order book, (3,5] indicates that the order was placed at depth four or five in the order book, and (>5) indicates that the order was placed beyond the top five prices in the order book.
|All values as % of total unique orders entered|
|‘Fast’ cancelled orders: Orders cancelled in less than 1 second|
|Market Cap Quartiles||At best||(1, 3]||(3, 5]||(>5]||Sum|
|Panel A: Spot|
|Panel B: SSF|
The value 2.47 in the first row of the table indicates that for the stocks with highest market capitalisation, 2.47% of orders were at the best prices, i.e. at the touch, and were rapidly cancelled. Similarly, the value 5.59 in the first row shows that for the highest market capitalisation stocks, 5.59% of the orders were at the best fourth or the fifth price level in the order book, and were rapidly cancelled. The last column adds up all the previous columns, and matches up with the share of orders cancelled within one second, which is Panel B of Table 1.
This table offers fascinating evidence about high frequency trading in India:
- The incidence of fleeting orders is very small: The biggest value seen is for Q3 stocks on the spot market, where 7.12% of orders were at the best prices and were cancelled within one second of their arrival.
- An overwhelming majority of fast order cancellations occur away from the best prices. As an example, in the 1st row, fast order cancellations were 36.83% of orders, of which 2.47 percentage points were at the touch.
- Stocks with the highest market capitalisation, where algorithmic trading is the most intense, experience a low incidence of fleeting orders as a share in total orders.
We cannot examine the intent behind these cancellations since it requires the knowledge of trader-identities for further analysis, which we do not have in our data.
Good regulation making requires data analysis and scientific evidence. The legislative function of regulators (i.e. the drafting of regulations) is primarily a research function: it requires deeply understanding the world, identifying market failures, and identifying parsimonious instruments of intervention that go to the root cause of the market failure. Globally, regulators such as the SEC and ASIC have deployed empirical research to determine the need for an intervention. The FSLRC handbook requires that regulators must do cost-benefit analysis before issuing any new regulation, where such research would be an early first stage of the regulation-making work. These capabilities are required in regulators in India if we are to build high performance organisations.
Our analysis above has many important implications for the policy analysis of the proposed minimum resting time of 500 ms:
- We used a more conservative measure — order cancellation within 1 s. We find little evidence of fleeting orders in India.
- We have undertaken the first stage of the research — counting fleeting orders. If the coercive power of the State were to be used in proscribing fleeting orders, SEBI needs to show evidence that this small proportion of fleeting orders is adversely affecting market quality.
- A regulation that interferes with all orders in order to influence the tiny proportion of fleeting orders is placing a burden upon society at large because it wishes to block a rare event. We should think more about the tradeoffs between prevention and enforcement. Perhaps it would be better for SEBI to build knowledge about how to enforce against market abuse in the HF environment, instead of imposing the costs of prevention upon society. SEBI’s proposal raises concerns about the possibility of faulty tradeoffs in security. There is a public choice theory problem here: It is a stroke of the pen for SEBI to impose restrictions upon citizens, while it is hard work for SEBI to build State capacity in enforcement.
SEBI’s discussion paper proposes seven interventions:
- Minimum resting time for orders
- Frequent batch auctions
- Random speed bumps of delays in order processing/matching
- Randomisation of orders received during a period (say 1-2 seconds)
- Maximum order message-to-trade ratio requirement
- Separate queues for colo orders and non-colo orders (2 queues)
- Restrict access to tick-by-tick data feed.
This article deals with the first: minimum resting time. As emphasised above, our work is limited: We have only counted fleeting orders, we have not gone into the question of demonstrating that fleeting orders have an adverse impact upon market quality. This kind of research is required on all the other six proposed interventions before policy decisions can be taken. This suggests the scale of research capabilities which are required before wielding the coercive power of the State in the legislative wing of a financial regulator.
The causal impact of algorithmic trading on market quality by Aggarwal N, Thomas S, 2014, IGIDR Working Paper.
The changing landscape of equity markets by Aggarwal N, Anand C, 10 July 2015, Ajay Shah’s blog.
Limit Orders and Volatility in a Hybrid Market: The Island ECN by Hasbrouck J and Saar G, 2002, Working Paper, New York University.
Technology and liquidity provision: The blurring of traditional definitions by Hasbrouck J, Saar G, 2009. Journal of Financial Markets, Volume 12, Issue 2, May 2009, p. 143-172.
Limit order revisions by Fong K and Liu W, 2010, Journal of Banking and Finance, Volume 34, Issue 8, August 2010, p. 1873-1885.
Limit Order Placement by High-Frequency Traders by Subrahmanyam A and Zheng H, Working Paper, 2016.