Creating a Quantitative Active Investment Strategy

In science, researchers start with an idea and set parameters within a study to learn if the idea is statistically significant. In quantitative investing, it's a similar approach. Where do you think quantitative investing starts?

But back to the hypothesis first—that's where it all starts. To begin the process, the active investor defines the market opportunity through public information to predict future returns of stocks. This is the investment thesis. Then it becomes all about data management. This can be a time-consuming process, as it takes time to build databases, map data, understand data availability, and find objective data sources. Plus, the data can come from a wide variety of sources and formats. What else do you think the investor must do within the data management step?

No. Just like a study, the quantitative approach begins with an idea that must be tested.

Incorrect. The quantitative method starts with an idea that requires proper authentication through testing.

Correct! The investor must clean and reshape the data so that it's consistent, error free, and usable. This process is no surprise when you look at the sources of data, like company mapping that tracks companies over time and across vendors; company fundamentals that include financials, demographics, and other market data; survey data such as corporate earnings, forecasts, and estimates by market professionals; and unconventional data such as unstructured data from satellites, news, supply chains, and corporate events.

No. That's going to remove data that could prove the hypothesis false.

Incorrect. By eliminating data that doesn't fit a required format, the investor loses key data that could help prove or disprove the investment thesis.

After the data is cleaned and organized, it's time to backtest the hypothesis. Essentially this is a simulation of real-time investing that's done over a historical period (usually 10 years). This process also includes rebalancing over a predetermined frequency to see how the strategy works. Then you'll get an idea of the factor's information coefficient or the correlation between factor exposures and their holding period returns for a cross section of securities. But there's a built-in assumption regarding the information coefficient in that it assumes an orderly relationship between expected returns and exposures. What's another way to put this relationship?

Correct. The information coefficient (IC) assumes that expected returns are linearly related to factor exposures so that it can have a positive correlation. The advantage of calculating the IC is that it takes information from all securities in the investment universe. The Pearson IC is the simple correlation coefficient (value between -1 and 1) between the factor scores for the current period and next period stock returns. Thus the higher the Pearson IC, the higher the predictive power for subsequent returns. In fact, any factor with an average monthly IC of 5% to 6% is considered very strong. In addition to the Pearson IC, the Spearman IC is basically the Pearson IC correlation coefficient between ranked factors scores and ranked forward returns.

Not quite. Factors should be positively related to expected returns.

Incorrect. Sporadic relationships won't be able be to forecast into a true relationship.

After using the IC, the next step is to create a multifactor model. This can be a complex task, since the manager can use their qualitative or systematic processes. Some researchers suggest that factors should be treated as an asset allocation decision, while others use a standard mean–variance optimization. Either way, once the portfolio is created, it's time to backtest the strategy, particularly with an out-of-sample backtest. However, the investor must be careful, as the backtest may not perform well in live trading. The manager usually calculates additional volatility measures to help understand the sample. What additional measures would be calculated?

Correct! The manager will calculate the *t*-statistic, Sharpe ratio, Sortino ratio, VaR, conditional VaR, and drawdown characteristics to better understand potential risks. This is an important step that's often overlooked. Many times, managers spend valuable time looking for successful models, but building the portfolio is just as important. Managers must also consider risk models that estimate the variance-covariance matrix of stock returns and explicit and implicit trading costs.

No. The backtesting sample already calculates the return measures.

Incorrect. Yield measures, like dividend yield, aren't going to help the investor understand volatility.

You'll also need to focus on some common issues in quantitative investing. First, survivorship bias can exist when stocks are excluded in the data that have left the investment universe due to bankruptcy, delisting, or acquisition. Essentially, this leads to an assumption that all investments will work out. What's another way to describe the effect of survivorship bias?

Yes! Survivorship bias is overly optimistic because it removes the negative securities that left the investment universe, so it can lead to wrong conclusions. Another major issue is look-ahead bias, where the investor uses information that was unknown or unavailable at the time an investment decision was made. This is essentially data mining, where financial data is used in a model that results in model overfitting. This leads to using excessive searching of past financial data to uncover patterns and conform results to something predetermined.

Incorrect. With survivorship bias, the investments are all perceived to work out well since the negative securities have all been removed.

No. It's not based upon average benchmark returns. Its impact is based upon positive returns, since the negative securities have been removed.

A final concern regarding active quantitative investment strategies is the side effects of using backtesting. In a backtesting environment, transaction costs, constraints on turnover, or limits on long and short positions aren't always able to be captured by the model's excess returns in a live trading period. Plus, some factor strategies, like short-term reversal strategies, will incur significant trading expenses as positions are created and then sold quickly.

Exactly. Quantitative active investment strategies start with a hypothesis that the investor believes in and then start collecting data and creating models. Once investors are confident in their model, the portfolio is constructed with risk controls.

To summarize: [[summary]]

With a hypothesis

With a created portfolio

With a desired rate of return

Clean and reshape the data

Remove all data that doesn't confirm the hypothesis

Eliminate data that doesn't fit the required format or structure

Linear relation

Inverse relation

Sporadic relation

Yield measures like dividend yield

Return measures like excess return

Risk measures like VaR and Sharpe ratios

It is overly optimistic

It is overly pessimistic

It is based on average benchmark returns

Continue