July 2017 Notes on Holiday Adjustment Methodology

The July 2017 meetup was lead by Chris Umphlett (LinkedIn), a Senior Operations Business Process Analyst at Jackson National Life. Chris discussed his research on developing models to predict staffing levels for days following a holiday for several departments at Jackson National Life. Chris has graciously shared the Holiday Adjustment Methodology summary used by Jackson National Life’s Operations department and the center of our discussion.

Not all holidays are observed the same way across the world or even within a county. This can be problematic when a company has offices in more than one country – or even more than one state – or is based in one country with a majority of its clients in another. These concerns lead to questions of how the company should schedule staff to accommodate their clients and associated workload. In some cases, international holidays are – sometimes – easier to schedule staffing because a majority of the world is celebrating the holiday but staffing related questions become murky when addressing national and local holidays. National holidays in the US are easily defined and are often referred to as bank holidays while local and regional “holidays” are more difficult. With respect to local and regional holidays it is important to take into consideration the lifestyles of the clients and employees.

Types of Holidays

There are different types of holidays but they can be categorized into the following broad types:

• International holidays (those observed in multiple countries)
  • Set dates (e.g., January 1^st, Christmas)
  • Moving dates (e.g., Ramadan, Passover, Easter, Rosh Hashana, Chanukah, and Lunar New Year)
• National holidays
  • Set dates (e.g., US Independence Day (Fourth of July), French National Day (Bastille Day; 14^th of July)
  • Moving dates (e.g., Martin Luther King Jr. Day, US Memorial Day, US and Canadian Labo(u)r day on the first Monday of September but May 1^st in other countries, Canadian Thanksgiving, and US Thanksgiving)
• Local and regional holidays
  • Mardi Gras (specifically New Orleans, Louisiana)
  • Sporting events (e.g., Super Bowl, The Final Four (basketball and hockey), and College Football bowl games)
  • Beginning of hunting season (region and state dependent)
  • Tax season (last week of January to April 15^th)

Type of Observance

When predicting the amount of staffing needed following a holiday the manner the holiday is observed impacts the amount of potential work. Holidays celebrated by a majority of clients impacts the “day after holiday” workloads differently than holidays celebrated by a small fraction of clients. Another consideration is the location of your clients. When the workforce and clients are in different countries – or even geographical regions with different customs – it is possible to have mismatched between available staff and projected workload.

• Mismatches between company-client holiday observation
  • Clients work/live in a foreign country and do not observe the holiday
  • Parent company is in a foreign country and does not observe the national/local holiday of the workforce
• The company does not observance the holiday but they experience an impact with work load and employee time off
• The company is closed for the day of the holiday
• Seasonal impacts
  • Retail purchases increase for upcoming events (start of school, holidays, etc…)
  • The beginning and end of a university’s/college’s semester/quarter can increase campus and town workload
• Depending on the field, the week(s) leading up to or following the holiday are abnormally busy or slow
  • The week between Christmas and January 1^st
  • Thanksgiving week
  • The Fourth of July (US Independence Day)

Date Consistency

When a holiday has a specific date – such as Christmas on December 25^th – it will fall on different days of the week. In the US, when the holiday falls on a weekend it is observed during the workweek. Thus, Christmas on a Saturday is observed on the preceding Friday and if falls on a Sunday it is observed on the following Monday. The problem of data consistency is demonstrated in the Holiday Adjustment Calculator PDF and illustrates how different methods to calculate the week of the year a holiday occurs results in divergent results. These concerns are summarized as:

• Same month and day (e.g., Fourth of July, Christmas)
  • Falls on Saturday, observed on preceding Friday (US)
  • Falls on Sunday, observed on following Monday (US)
• i^th weekday of the month (e.g., Thanksgiving (Canada and US))
• Based on a lunar or non-Gregorian calendar

Other Problems Associated with Holidays

In addition to knowing when the holiday occurs there are concerns regarding staff availability, fiscal calendars, number of business days in the month, and the number of weeks in the year.

• Week of the year/quarter/month concerns
  • Which day does the year/quarter/month start?
  • Taking into consideration if there are partial weeks
  • How are the weeks being counted? Does the year/quarter/month start on week 0 (zero) or 1?
  • Are there 52 or 53 weeks in the year?
  • How many Fridays are in each month?
• Number of business days in the week/month
  • Partial week?
  • Reduced number of days in the month to accomplish the same amount of work

Also discussed…

• Text mining and word association network graphs to group people with various skills. The package tidytext(GitHub and CRAN) is a great way to mine and analyze text; check out the book and the vignette. The UC Business Analytics R tutorial on Text Mining: Word Relationships demonstrates how to create network maps between related words.
• Notebooks for sharing analytics workflows:
  • Apache Zeppeline: A multi-purpose, web-based notebook for data analysis and collaboration able to support many interpreters (R, Python, Apache Spark, Markdown, shell, SQL, and Hadoop to name a few).
  • IPython notebooks: Official website and Project Jupyter
  • R Markdown Reproducible Reporting: Webinar from RStudio.
  • R Notebook Workflows: Course from rstudio::conf2017
• ggplot2 and ggplot2-esque packages
  • All Your Figure Are Belong To Us provides examples of various R packages for creating very specific plots.
  • The GGally R package (CRAN, GitHub, and examples) provides a function to create graphical correlation matrices using ggplot2.
  • The ggplot2 extensions website is a collection of R packages that extend ggplot2 to make animated plots, change and modify the look, create radar plots, htmlwidget to make interactive plots, keep labels from overlapping, visualization and annotation of phylogenetic trees, and construct networks and trees.
• Technical debt related to writing packages with documentation and examples for collections of project related functions; see here for more information.
  • Creating packages in RStudio is a great introduction along with Hadley Wickham’s book R packages to creating R packages.
  • Make sure you write unit tests for your functions. Hosting your R package on GitHub? Use Travis CI to check the integrity of the package; Travis CI getting started guide.
  • 97 Things Every Programmer Should Know edited by Kevlin Henney is an excellent book on being a better programmer that will lead you to being a better citizen along with writing better code and applications.

back