Are AI and Machine Learning the Key to Understanding the U.S. Economy?

GPUs fuel AI and machine learning. Initially created for video games, they are used in sports and business analysis by fantasy baseball enthusiasts, oddsmakers, and front office executives who want to enhance their understanding of the hidden value of often obscure players. Other uses of this technology’s extreme processing power include the recognition of animals, such as dog breeds or endangered species, to allow biologists to gain a more accurate understanding of species populations in a geographical area.

GPUs and advanced statistics constitute an incredible advancement in 21st-century technology and can result in a more precise and deeper understanding of the data that is captured by a plethora of sources. So, why is so much of the economic data used by the news media and the U.S. Bureau of Labor Statistics for unemployment or gross domestic product (GDP) analysis still the same as what was used decades ago?

Data for GDP and Unemployment Analysis

The initial concept of the GDP (originally referred to as gross national product [GNP] and calculated slightly differently than GDP) was first conceptualized in the 17th century. The GDP measures the value of all goods and services produced within a country’s border within a specific point in time. The modern version of the GDP was developed in 1934 by Simon Kuznets for a U.S. Congress report to measure the U.S. recovery from the Great Depression. Kuznets’ work ultimately resulted in a Nobel Prize in Economics.

By comparing the GDP with prior periods, it can be determined if the economy is expanding by producing more goods and services or if it is contracting (two consecutive quarters of negative growth constitutes a recession). It could be argued that the GDP is the most critical indicator of the health of a nation’s economy, yet little has changed in how the U.S. measures GDP since it was first utilized in 1934.

The second most-watched indicator is the unemployment rate. The GDP is calculated by the U.S. Department of Commerce. Since 1929, the unemployment rate has been measured by a division of the Department of Labor. You would think the number represents the number of people collecting unemployment insurance, or, better yet, the actual percentage of people not employed or underemployed. But instead, the government does a random sampling of 60,000 households. Based on the results of the survey, the unemployment rate is then derived. The unemployment rate purposefully does not include “discouraged workers,” or workers who should be retired but are still looking for work, or those not eligible for some reason to collect unemployment insurance.

Modern Analysis Requires Modern Data Capture

The greatest omission of the unemployment metric is the massive distinction that has developed over the last 100 years regarding the definition of being employed. A century ago, the concept of a job was quite straightforward and binary. A person was either employed or not. There was no need for the great majority of employed people to have multiple jobs. Families could easily subsist on a single job. However, in the 21st century, there are countless variations of employment situations. The purveyor of a shared ride company who is self-employed and works 20 hours a week is considered employed, as is the middle-aged grocery store worker, yet neither can manage a life, with or without dependents, on the income generated from such a job.

Currently, the unemployment rate is low (as of early 2020), but this means very little. It is, however, true that both the old processes for calculating GDP and unemployment are considered to be “leading indicators,” acting as a proverbial crystal ball for those predicting global economic health.

Policymakers in the government use this information to make decisions on programs that affect every citizen’s life. Using this information, the Federal Reserve Bank raises or lowers interest rates, impacting everything from the cost of home mortgages to the cost of capital needed by companies for investment purposes. Investors use these numbers to determine where they might invest or how they might allocate their investments. However,  little has changed in close to 90 years with regard to how the GDP or unemployment rate is calculated.

Would you feel confident that the CEO of a modern corporation was acting in the best interests of shareholders and customers if the organization only made decisions based on the same two datapoints it had available to it in the 1930s? Could a company be competitive or reach its full potential by ignoring the ocean of data available to it? Given the complexity and size of the U.S., are the news media and government acting in everyone’s best interests by relying so heavily on the unemployment rate and GNP? Yet, today, these two numbers are the foundation on which many of these decisions are made.

For years, academia, the news media, and the U.S. government have been fixated on collecting data on all facets of business and society, yet they are long overdue in implementing the latest technologies to put all this data to good use.

Just as many a modern corporation uses key performance indicators to measure how effectively a company is achieving critical business objectives, there is an effort underway to create a national indicator system for the U.S. The State of the USA (www.stateoftheusa.org) was established to gather public and private data and make it available.

Anyone who visits the catalog of U.S. Government Publications (https://catalog.gpo.gov) will see that there is no shortage of data. Just as in the private sector, the variety and velocity of data are growing at an alarming rate. Turning all that data into actionable insight requires applying 21st-century technology.

Time to Update Economic Algorithms

We are at a turning point in history. Machine learning and AI could hold the keys to understanding every lever that impacts the economy and help us predict what will happen if and when those levers are applied.

Modern GPUs have made the dream of computer scientists a reality. AI using machine learning models driven by nearly infinite amounts of data can be utilized to derive results of unparalleled insight. An average minor league baseball scout has access to large quantities of useful data on an 18-year-old right-handed pitcher, such as the probability of injury based on dozens of different physiological factors derived from thousands of hours of film of the pitcher throwing. That same scout has access to a plethora of other valuable data and can set tunable parameters to shift the probability of successfully finding the next Gerrit Cole, who just signed a $324 million contract with the New York Yankees, as opposed to the next Tommy John surgery candidate.

The same scout has indispensable tools immediately available on every obscure player under consideration. But, so do the fantasy baseball companies, as well as the gambling organizations and the simple hobbyists.

So, we end with a simple question: Why aren’t academics, the news media, and the Bureau of Labor Statistics taking the same approach when calculating economic metrics? Why does a part-time, self-employed Uber driver count as being “employed”? Why is it that the economic algorithms fail to utilize the same level of sophistication and technology used by fantasy baseball?

Image by Gerd Altmann from Pixabay



Newsletters

Subscribe to Big Data Quarterly E-Edition