Authored by: Erez Katz

Introduction

On a recent Sunday morning news show, former Obama Chief of Staff and Mayor of Chicago, Rahm Emanuel, reprised his famous quote from the 2008 financial crisis, “Never let a crisis go to waste.”  For investment professionals there is no better time to realize the transformational power of an algorithmic investment approach powered by machine learning and alternative data.

In today’s blog, I want to showcase a strategy which targets the retail and automobile sectors, both severely impacted by the Covid-19 lockdown.

My goal is to succinctly present how we were able to stitch together multiple data sets and dynamically train machine learning models to inform when to move in and out of positions in a high-volatility and emotionally charged environment.

Dynamically Trained Retail and Automobile Strategy Using Equifax Consumer Credit Data

Equifax Retail and Auto Dynamic is a strategy predicated on Equifax’s aggregated and anonymized consumer credit trend data. The strategy selects optimal entry and exit for large cap retailers (including auto retailers) and holds them as long as they satisfy the conditions that selected them in the first place.

The AI engine retrains its models repeatedly and dynamically identifies which combination of Equifax, fundamental, and technical  features are most informative for the current market regime. The strategy selects its constituents using a two-phased screening process. First, it identifies a high-conviction entry at the sector level using Equifax’s trend data. It then further screens for constituents exhibiting strong fundamental/technical conditions relative to their peers.The strategy focuses mainly on 150 or so dominant retailers.

Image 1: Backtest of  Equifax Retail and Auto Dynamic against XRT (SPDR Retail ETF). For a full backtest report click here. Past performance is not indicative of future returns.

The above performance report is completely out-of-sample. In addition, it accounts for transaction costs and slippage. As you can see from the strategy’s performance (in orange), the strategy only acts on high-conviction entries. In other words, it’s perfectly happy staying in cash for a prolonged period of time. Conversely, it’s equally happy to stay in a position in order to fully realize its upside potential. This particular strategy exhibits low turnover with over 70% successful benchmark relative transactions.

For the astute investor, the step formation (in orange) and the muted volatility (see max drawdown of 9.8% vs. 47.02% of the benchmark), are exciting evidence of the strategy’s potency. A hedge fund trader can easily leverage it 5X and it will still be on par with the benchmark’s volatility (XRT, SPDR Retailers ETC).

Focusing on the most recent market regime (see chart below), February 2020 to present, the strategy was able to stay in cash during the big selloff in March and also was able to jump back in gradually as the market started to level off and revert back from the massive selloffs.

Image 2: The most recent 3 months of Equifax Retail and Auto Dynamic against XRT (SPDR Retail ETF). For a full backtest report click here. Past performance is not indicative of future returns.

The green histogram representing the number of constituents in the portfolio (see the bottom of the image), shows the portfolio was mainly in cash during March and into early April, after which it started  to introduce new constituents into May 8th (last Friday).

Dynamic Feature Selection and Model Retraining

It’s important to note that the Equifax Retail and Auto Dynamic strategy is focused on retraining its models based on the most recent market regime. It uses a concept we call the “roll forward retraining window. Our software development team has recently added full transparency of the underlying models used in our  backtest. The strategy’s performance report clearly delineates which features are used during their respective time period.

You can clearly see how credit default rate, or credit utilization are used in conjunction with stock-specific factors such as EBITDA, or debt-to-income ratio.

If you hover over the icons in the live report here, in the Dynamic Model Timeline section, you can clearly see which models (and their respective factors) were used at various time frames.

Image 3: The Dynamic Model Timeline section which describes which factors and their respective thresholds were used each period of time. For a full backtest report click here. Past performance is not indicative of future returns.

Conclusion 

In today’s blog, I wanted to provide a clear example of how Equifax consumer credit data can be used effectively to navigate even in the most unpredictable times. The strategy depicted here exemplifies the power of alternative data coupled with sophisticated AI technology. By employing an algorithmic, discretion-free approach to investment strategies such as the one described, investors can make their decisions solely based on data and statistics, fully exploiting other investors behaving irrationally.

If you’d like to learn more, or if you ‘d like to sign up for a trial for Equifax smart data feeds or Lucena’s model portfolios, please feel free to reach out to me directly.

Lastly, next week is retail data week. Our expectations are that the numbers will  be better than the street is expecting, given the bullish posture of our model portfolio.

Have a profitable week!

Questions about algorithmic trading using alternative data? Drop them below or contact us.