Select Page

This is part of an ongoing series on the multi-step approach to qunatitive trading. Be sure to check out part one, Identifying a Strategy.

Once you’ve identified a solid strategy, you just can’t put it into place. You will need to run a simulation in a live environment and, even more importantly, see how it fared in the past. In other words, you would probably want to backtest your strategy over historical data. While good performance against the past does not guarantee equally good future performance, it still provides a solid forecast since the data is mined and real.

While not perfect, historical backtesting provides a better picture of expectations. It is a very complex component on its own, first and foremost due to the sheer quantity of data, which is also not easily accessible. Additionally, running the simulations may take hours upon hours.

What We Mean By “Non-Accessible Data”

If you’re trading from a mobile app, you’re probably basing your decisions off of past performance, newsworthy updates, and end of day statistics. Those are easily accessible markers, and useful in some instances. But when you’re using algorithms to facilitate lightning-fast trades, your data is far larger in scale and complexities.

A few of the changes include the following. First, you may need access to tick data. Tick data refers to any market data, which shows every price and volume throughout the trading day, often with thousands and even tens of thousands of updates for every traded security. Additionally, tick data often includes information about every change to the best bid and offer of the various buyers and sellers throughout the day. There can even be hundreds of thousands of updates for each stock.

Second, stocks often trade in many different venues— such as NYSE, Nasdaq— and you will need data from all these venues. Third, you will need to accurately time stamp the data events and make sure the data arriving from different venues is adequately synchronized. These are only a short list of some of the primary challenges.

In sum, quantitative trading kicks it up a notch by tracking down all the potential data available— and that goes well beyond the end of day stats or financial news you’ll listen to on your commute home. In addition to the usual markers, our programs analyze every time a stock is traded, posted, or otherwise offered to buy or sell. Each day, data of this variety can easily add up to be tens of thousands of gigabytes. Just the sheer to direct market data costs around millions of dollars a year as of the data of the post. Even then, the data might not be as good as it can be. For example, how can you be sure the compiler synchronized all of the day’s trading events with accurate time stamps?

In order to ensure accuracy, qSpark records its own data. With direct lines to all exchanges in North America, we can channel all of that data into centralized, colocated servers. On the floor, our  machines are recording data from all of the exchanges.

Next time, we’ll be looking at strategy execution.