 "Backtesting of algos soon to become norm at exchanges")
In a first, the country's two premier bourses, the National Stock Exchange of India (NSE) and the BSE, are gearing up to introduce measures to back-test algorithm programmes, or algos, going back as much as five to 10 years. The aim is to test if the algos can perform normally during bouts of volatility and during events that resemble black swans like the collapse of the Lehman Brothers in 2008 that triggered the global financial crisis.
The move comes in the backdrop of the market regulator’s plans to impose curbs on algorithmic trading. “We believe that the back-testing framework is a useful mechanism to stress test trading algos against extreme events like abrupt market movements, and the outputs can be used to refine the algos so as to avoid errors in live trading, making markets a safer place,” said Kunal Nandwani, chief executive officer, uTrade Solutions.
An e-mail to NSE and BSE, asking them to elaborate on the testing mechanism they plan to put in place remained unanswered. The framework is expected to be ready within six to nine months, sources said.
At present, exchanges do not back-test algos but do mock tests to check how these will behave in the live market. Algos have to go through 13 checks before they are given an exchange's approval, after which the strategy is allowed to go live.
This includes checks on span and order limit size, and tests to ensure the algo does not go into a loop. A span check ensures there is enough amount in the brokers' account before an algo is fired, while the order limit restricts the order value to Rs 2 crore. “What exchanges are most bothered about are algos going into a loop, wherein an unlimited number of orders are fired instantaneously,” said Nithin Kamath, founder, Zerodha, a discount broker.
According to Kamath, the problem with mock tests is that they are done with fake ticks and there is no way of knowing how the algo will behave in the live market. “In a mock test, only 200-300 participants trade with thin volumes, which is not a good way of analysing the algo. While back-testing will prove beneficial, exchanges need to find a way to simulate the real market,” he said.
Recently, the Association of National Exchanges Members of India had written to the Securities and Exchange Board of India, asking for an increase in risk layers to manage the interaction between participants and exchanges, including controls on the number of messages per second, the number of orders per second and maximum open orders for a product.
Algo-based trades use advanced mathematical models for transactions and can pump in thousands of orders in a second. While such trades provide liquidity, as more orders are placed, these could distort prices if wrong programmes are allowed to run unchecked.
As of June, 18 per cent of the turnover on exchanges came through the algo route. Another 24 per cent was accounted for by co-located servers.
TESTING TIMES
- Exchanges gearing up to introduce measures to back-test algos going back as much as 5-10 years
- The move comes in the backdrop of the market regulator’s plans to impose curbs on algorithmic trading
- Exchanges do not back-test algos but do mock tests to check how the algos will behave in the live market
- Algos have to go through 13 checks before they are allowed to go live
- Exhanges are most concerned about are algos going into a loop, wherein an unlimited number of orders are fired
