Finding Predictive Insights In Consumer Transaction Data

  • By Eric Davidson
  • March, 01
Blog Finding Predictive Insights In Consumer Transaction Data

Quantitative investment portfolios, also known as systematic or algorithmic portfolios, have many advantages, such as the ability to scientifically evaluate dozens of factors in near real-time and execute orders without emotion. While constructing quantitative portfolios involves algorithms, data and technology – the process doesn’t have to be a black box. Algorithmic investing strategies leverage algorithms and data-driven techniques that are initially researched and tested by data scientists and quantitative researchers – people who are trained to identify actionable signals but also trained to ensure that signals have a solid economic rationale (i.e. explainable).  As long as Portfolio Managers and Analysts leverage such data and given the transparency in the processes, the notion of a blackbox is dispelled and subsequently leads to faster and broader adoption by both investment professionals and regulators.

This article provides an insight into how we use technology and human judgment to unlock competitive advantages in consumer transaction data – in a transparent way.

Constructing a Thematic Portfolio Workflow

Neuravest has developed a scientific approach to bringing thematic portfolios to market.

Idea generation and Economic Rationale: It starts with an idea or thesis generated by a subject matter expert. Ideas generally involve events, catalysts, or information sources that the market hasn’t discounted and that may affect forward returns. Typically, thesis statements begin with a sentence such as, “Transaction data from credit cards and debit cards can provide detailed information about the performance of certain companies before the market has fully discounted the information. By leveraging this informational advantage, we can construct portfolios tilted towards companies set to outperform industry peers or a benchmark.”

Data Identification and Validation: Researchers will then aggregate the necessary alternative datasets (transaction card data) and traditional datasets (sell side analyst consensus data, stock prices) required to validate or invalidate the thesis. The Neuravest Data Analytics Suite (DAS) platform supports a comprehensive workflow by which we determine if a dataset is indeed predictive and conducive to the strategy’s objectives.

Model Building, Portfolio Construction and Back testing: A team of investment professionals then engineers features, performs advanced statistical analysis and then optimizes portfolios with in-sample, out-of-sample and walk forward simulations.

Production: Once the portfolios have been through an exhaustive validation and adjustment process, including paper trading simulations, we’re ready to deploy capital (initially small) and increase exposure with cross-validation in new market regimes.

 

Neuravest approach to systematic portfolio construction

Image 1: Thematic portfolio workflow

 

Data Source

Neuravest has partnered with 90 West Data, a company with an exclusive panel of US consumer transaction data. Neuravest receives US debit card and ACH transaction data that covers multiple dimensions of consumer spending. The data is provided daily with a two-day lag. The infographic below illustrates this concept. 

Lags in Credit Card Data versus Lag relative to Earnings Release

 

 

Corporate earnings announcements occur quarterly with up to a six-week lag. Thus, data from debit card and ACH transactions may provide an early indication of earnings growth if the datasets are positively correlated – an information advantage.

 

Case Studies

To determine if 90 West’s transaction data is positively correlated to actual revenue growth, Neuravest researchers selected a group of consumers who have used their debit cards to spend at Starbucks and Wayfair in the current and recent quarters. Below we can see this spending is highly correlated to actual revenue growth reported by the company in official quarterly earnings releases.

Starbucks (SBUX)

 

Wayfair (W)

Further, changes in average ticket size on normalized consumer spending panels may provide insight as to whether revenue growth is being driven by price or number of transactions.

 

Caveats

It’s important to note that credit and debit card data must be normalized and adjusted for seasonality as consumer spending patterns are different at varying times of the year. Additionally, investment signals from transaction card data are most effective for consumer sectors (“B2C”)– as opposed to companies that sell to businesses (“B2B”). The revenue metrics of consumer-facing companies are highly correlated to credit and debit card transactions.   

The graph above depicts how de-seasonalized time plots may be useful in forecasting future trends because sharp seasonal peaks and troughs are smoothed out, providing more basic trend information.

 

Feature Engineering

Given the strong evidence of predictive information in 90 West’s data, Neuravest conducted back tests on whether features derived from the dataset can be fed into a machine learning classifier. (A classifier is a machine learning objective which targets a predetermined state.) Neuravest has found Growth Acceleration/Spending per Transaction to be an especially important feature for predicting ex-post price returns.

Neuravest research objectives are two-fold:

         a.  Forecast a company’s future revenue growth and compare it to the revenue growth from sell-side analyst estimates – identifying any meaningful variance by comparison.

         b.  Identify abnormal changes in consumer spending and determine if associated characteristics are ex-ante predictors of abnormal ex-post price changes.

 

What’s Next?

Please see the following link for additional detail around derived portfolio features and signal drivers in a White Paper, including processes for in and out of sample validation – as well as how we move the portfolio into production.  For White Paper Click Here

Conclusion

Embedding artificial intelligence and data science tools into portfolio management is an expensive and time-consuming undertaking. Yet, the effort may have a payoff. According to Accenture, firms with AI and signal generation capabilities are seeing better alpha generation, better portfolio allocation and better portfolio construction.[1] Neuravest offers a more efficient, less capital-intensive solution for asset managers who wish to work with alternative data sources and generate new validated portfolios.

We showed how insights in consumer transaction data from 90 West can be extracted in a transparent way. The economic rationale underpinning the use of debit card data is intuitive – as consumer spending represents a form of ground truth data.  Further, consumer transaction data gives us broad insight into consumer spending patterns at the company level and broader retail sector level. The data is updated more frequently than official company data releases, offering quantitative investors a potential information advantage.

Through our strategic data partnerships with 90 West, Neuravest licenses thematic portfolios to institutional investors seeking equity alpha. Investors benefit from cost-efficient idiosyncratic factor exposures in a single, separately managed account. Additionally, Neuravest can work closely with institutional asset managers to build an investment portfolio with 90 West’s data that is unique to their firm and risk mandate – optimizing for specific dynamics such as portfolio turnover and volatility constraints.

For more information, please contact: eric.davidson@neuravest.net

[1] https://www.institutionalinvestor.com/article/b1qlgbj7rstm2r/Artificial-Intelligence-Isn-t-Just-for-Quants-Anymore

Have a media inquiry or a topic you’d like to contribute to our blog?