Category Archives: Greg Robin Random Walk

Promotional Index Launch

With more than 6 years of historical data tracking email promotional activity from nearly 300 consumer brands we are excited to launch our first data product!

This fall select clients can access structured weekly data for this first time.

Investors will be able to quantify and help answer the following:

  • Changes in campaign frequency
    • Did a brand run more marketing campaigns in order to drive sales?
    • Did a brand reduce campaigns as demand increases?
  • Growth in customer and lead lists
    • Has a business grown its lead list?
  • Intensity of promotional offers and discounts
    • Did the size of discount increase as a seasonal sale progressed?
    • Did the coupon specify a higher discount than prior periods?
  • Changes in promoted products
    • Are there changes in which products are being discounted?
  • Changes in targeting strategy
    • Is a brand struggling to hit sales quotas engaging in more “blasts” to its entire list?
Compare current promotional cadence with prior periods

90% Returns: Random Walk Portfolio Utilizing Promotional Intelligence Outperforms Benchmark

Random Walk partnered with Lucena Research to better understand the impact of email intelligence on share prices.

Lucena Research conducted a robust nearly 3 year backtest comparing the retail benchmark (XRT) with Random Walk’s(RW) proprietary methodology measuring email promotional aggressiveness. While the XRT returns were nearly flat, the RW portfolio generated 90% returns.

The RW Model uses Lucena’s machine learning to incorporate RW factors such as declines in promotional activity, compares steep discounting email volumes and overall promotional email volumes to determine portfolio constituents.

Here is how the Backtest works:

  • Each day in backtesting the period from Dec 30, 2016 to July 23, 2019 we scan the Random Walk Universe (about 150 consumer stocks) for constituents that match the criteria for a long entry, which is determined during the training period from July 1, 2015 to December 12, 2016 using Lucena’s proprietary Machine Learning Algorithms.
  • The long position is held as long as the model tells us to.
  • The minimum value for allocation of each constituent is 5% and max value is 25% of the portfolio.
  • Transaction cost and slippage are considered in this backtest.

Intuitive analysis of the signals generating events:

A long event is selected when:

  • 12 week moving average of emails sent after log normalization and ranking against peers with discount percentage of 51- 100% is lower than 0.25 or in simpler terms when the number of emails sent in higher discount category of 51-100% is less over the previous 12 weeks as compared to other consumer brands.
  • 18 week moving average of projected total volume of emails sent after log normalization and ranking against peers is higher or >0.75, or in simpler terms, when larger numbers of emails are being sent by the company over the previous 18 weeks compared to other consumer brands.
  • Ratio of Trailing Twelve Month Earnings to Market Cap ranked against Russell 1000 constituents is between 0.6 and 0.971.
  • Volatility over previous 252 days ranked against Russell 1000 constituents is between 0.02 to 0.6.

Transactional Data: Institutional Investors’ Fool’s Gold

Why Credit Card Data Fails

Big data in investing is here and we are a part of it. However, some institutional investors are getting confused as to the end zone.

There are several expensive credit card transaction products that correlate well with coincident revenues some of the time. With Wall Street investors trained to drool at regressions, error bands and correlations, the allure that these products can do the decision making is appealing. While often accurate in predicting some component of revenues, this “data” approach has herded investors into a series of grossly inaccurate conclusions on consumer stocks in 2017.

To be specific, the largest big-data transaction vendors led investors right to slaughter in a wide range of mall based retailers this spring. Ironically, the data generally projected somewhat accurate revenues in names including JC Penny (JCP), Macy’s (M),  Foot Locker (FL), Michael Kors (KORS), Vitamin Shoppe(VSI) and several others. In summary, the transactional data indicated results were to be inline, an expensive and grossly wrong conclusion.

Why did this expensive credit card data fail miserably? Look how tight those error bands are! The transactional data provided no context into organic demand of these products. Credit card receipts did not factor in the desperation among retailers engaged in their steepest discounting in history. Share prices collapsed and investors overly focused on transactional data were left with a classic Pyrrhic victory, attempting to take solace with the mantra of “but we were right on revenues”.

Other times, the transactional data is correct most of the time when its business as usual, then misses the largest move because something unusual has occurred the sample did not detect. So investors are correct, when there is no money to be made, and wrong right when a business is about to be dramatically revalued due to a massive inflection in demand. Vitamin Shoppe (VSI) comes to mind in this scenario. The Random Walk ensemble captured the 50% plus collapse in business as customers migrated to Amazon and stopped by placebo pills altogether. Through a combination of click stream data, review volumes, email responses our more robust, less precise approach generated alpha for our clients.

In terms of risk reward, compounding the problem is the herding mechanism related to this mass consumption transactional data. Every highly transactional, well resourced hedge fund is viewing the identical data sets, yielding the same conclusion. This now incorrect herd exacerbates the share price collapse as panicked analysts- now uncommitted shareholders all attempt to rush through a tiny exit door at once.

In contrast, our more robust and diverse data ensemble can be less precise for coincident revenues, but more ACCURATE in predicting changes in consumer behavior that ultimately influence share price. Our view is that new customer acquisition volumes, organic demand for products and repeat customer frequency drive share price. These are the metrics we focus on predicting because the relate future growth. We don’t want to predict the present or past.

Our data ensemble detects Blue Apron giving away food to attempt to bring back past customers.