Alternative Data Roadmap: Five Below

Introduction

This outline shows how a variety of alternative datasets could be used to gain a better understanding of the retailer Five Below. It will review a number of KPIs (Net Sales, etc.) sourced from the latest earnings call and 10-Q filing, followed by related alternative data sources and strategy. Analysis for each KPI is generally thought of as having two objectives: Validate and Project. When a KPI is stated by management, the first step is to Validate the alternative dataset through backtesting. Once a high degree of correlation is established, we can then Project the KPI for future quarters. These projections can then be used in the underlying financial analysis to value the company.

After reading this roadmap, check out epicurusdata.com for a sample backtest of FIVE using web traffic data (shown in image above).

Summary

The main KPIs driving FIVE’s business are growth in comparable sales and growth in new locations. A full list of existing and planned store locations is crucial to understanding the growth potential of the company, and from a data perspective is required to derive a comparable sales number.

The following datasets would deliver a solid initial picture of Five Below’s business:

·     Geo-Location foot traffic, such as Cuebiq, Thasos.

·     Credit/Debit Card, such as Yodlee, 1010.

·     Email Receipt, such as Slice/Rakuten, Superfly.

KPIs

Below are the main KPIs from the earnings call and 10-Q filing.

Meta Data

A good data project generally requires high quality meta data as an input. There are some data vendors that will pre-map the data before selling it, but there are advantages to creating a proprietary mapping: control over quality, transparency when an issue arises, and most importantly proprietary insights.

The primary mapping for FIVE is store location to lat/long, or ideally a geofence. AggData provides a latest-snapshot dataset with locations for thousands of companies, but lacks historical changes. To supplement this data, you could scrape Five Below’s website “new locations” section, combined with Wayback Machine snapshots, to recreate a historical timeline of locations.

Comparable Sales

Building a comparable sales estimate from Location Data first requires dividing stores into Comparable/New for each quarter historically, using the definition provided by the company on pg. 16 of the 2018Q1 10-Q filing. With this mapping, foot traffic can be panel adjusted and aggregated up into quarterly indexes. Next, data can be Validated against previous reported quarterly numbers; this ensures the dataset has predictive value. Finally, the data can Project the comparable sales, new store sales, and any derivative metrics such as average sales per store or regional sales totals.

Product Level Sales

Five Below reports 3 product segments (Leisure, Fashion and Home, and Party and Snack), and focuses on 2 derived metrics (Basket, Transaction count) for analyzing the source of sales growth. The sales can be further divided into 2 channels: brick and mortar and ecommerce, which correspond to the vendor types for alternative data.

Slice/Rakuten provides item-level descriptions for ecommerce sales, pre-mapped to ticker. Once the products are bucketed according to the 3 segments above, sales figures can be calculated by aggregating and panel adjusting. Once again, after Validating the data against previously reported figures, Projections can be made for product segment sales. The same can be done for basket (items per receipt) and total transaction counts.

As for brick-and-mortar sales, alternative datasets are generally divided into electronic payment such as debit/credit cards, and cash. Electronic payments are easier to track, so for an initial model the cash transactions can be assumed to follow the same trends as electronic payment. Debit/Credit card datasets provide user-level transactions, but do not provide insight into the products a user purchases. For this reason, this dataset type is useful for transaction volume projection but not basket or segment sales. However, retailers often label transactions with a store identifier or location – if this is the case for FIVE, debit/credit card data can be used to duplicate the KPI projections from the location data (comparable/new store sales). The same technical process applies here, namely map/aggregate/panel adjust to Validate, then Project the KPIs.

Conclusion

This roadmap has gone over an initial modeling process using alternative data to project a few KPIs of high importance, which in turn are used in a valuation model. Additional dataset types such as web traffic to checkout pages, cash receipts, etc. can and should be used to further validate projections for KPIs. The “Unique Metrics” section of the KPIs page describes a number of other metrics that are seen as impactful either by management of FIVE or by financial analysts asking questions on a conference call. Each of these can be modeled using other alternative data sources – weather databases, twitter feeds, TV ad records, etc… but sometimes simplicity is key!  

About

Ben Griffin is a financial professional with 4 years of experience in alternative data and 7 years in financial technology. Currently based in Brooklyn, NY his primary focus is in the union of data engineering and analytics in a better understanding of financial analysis. Open to contract work, partnerships, or full time employment. Please reach out to ben.griffin23@gmail.com for inquiries.