The rapid growth of algorithmic trading in Canada — which is expected to continue unabated for some time, particularly for trade efficiency and productivity reasons — presents a number of benefits and challenges to real-time market regulation.

One challenge is effectively monitoring the sheer volume of orders and trades that algorithms generate. More than 320,000 transactions are made daily on Canadian exchanges and alternative trading systems. For each trade, there may be more than 10 orders.

The increased order flow resulting from algorithmic trading, which uses sophisticated mathematical formulas to identify and execute “best trade” opportunities, means more order changes and cancellations, and multiple messages for our systems to process.

To remain effective regulators, we have to ensure that our surveillance systems have enough capacity to monitor this increased order flow in real time.

Algorithmic trading also changes the nature of the regulators’ work by making some tasks more complex and others less onerous. Required skill sets are changing.

Today’s algorithms vary from simple participation algorithms, which split volumes into small amounts, to more complex algorithms, which respond to observed liquidity and try to obscure their nature.

The simpler participation algorithms are easier to see. Their trading patterns are noticeably different from more typical types of trading. However, some of the more complex algorithms are harder to see and, in fact, are designed to avoid detection.

There is nothing nefarious here, but it makes it more difficult to understand what is really happening in the market or the true intentions. And intent is a key factor in determining whether a particular activity or trading behaviour is manipulative or not.

From the dealer’s compliance perspective, algorithmic trading requires careful performance monitoring as well as post-trade analysis to ensure it is properly applied and used for best execution. Dealers offering algorithmic trading also retain compliance responsibilities for their clients’ trading. Dealers must ensure that the orders entered — whether by a machine or a human — are not deceptive, don’t create artificial prices or volume, and don’t violate client priority rules.

Algorithms are programs, and programs can have errors. If an algorithm isn’t programmed correctly or isn’t getting timely information, the client can have significant exposure. And, if the dealer knows a client is using a machine to generate orders automatically based on order and trade pricing on multiple marketplaces, the dealer must be satisfied the algorithm is generating orders according to the trading rules, rules of the marketplace and the firm’s limit for exposure.

Risks can arise when a trader pursues algorithmic and non-algorithmic strategies that may interact in ways that trigger anomalies and alerts, if not violations. If the dealer, trading as a principal using an algorithm, reacts to a pattern of trading caused by a client’s algorithm, this may spark alerts or inadvertent best execution and client priority violations.

Conversely, algorithmic trading can make our work less onerous. Very liquid markets with tight spreads and depth in the book provide the most effective price discovery and are the fairest markets, as they are characterized by low volatility. When the order book is lined by a large number of orders generated by participation algorithms, spreads are reduced and depth increases.

This means you won’t get the price swings you’d get if those orders were not there. Adding orders to the book makes it easier to monitor the market because there is less volatility from the reduced probability of unexpected significant price movements.

As the market becomes more liquid, our surveillance system generates fewer price and volume alerts, freeing up staff to focus on other issues, such as rule-based alerts.

The increase in visible orders from algorithmic trading benefits those who trade, not just those who place orders. As algorithmic orders narrow spreads and increase depth, each order that actually trades gets a better price than if the book was not so deep.

Intermarket algorithms that exploit price differences may also reduce the impact of trade-throughs. If a number of algorithmic systems generate orders to take advantage of price differences between marketplaces, price levels will eventually start to converge.

This greatly reduces the risk of significant trade-throughs because marketplaces will have more uniform trading prices and the remaining differences in prices between marketplaces will probably be accounted for in transaction costs.

As volume and speed increases, our real-time surveillance focus will broaden to incorporate more near-time surveillance techniques, such as data mining, using our own algorithms. IE

@page_break@
Tom Atkinson is president and CEO of Market Regulation Services Inc., the independent regulator of equities trading in Canada. This is an extract from his speech to the first Canadian conference on algorithmic trading in Toronto in February, co-sponsored by RS and TSX Markets Inc. Gerry Rocchi, an RS board member, contributed to the speech.