Search This Blog

Featured Post

Introduction

Hello all readers, Welcome to The Parmeter Politics and Policy Record ! My name is Nathan Parmeter, an aspiring public policy professional a...

July 02, 2020

Quantitative Modeling of Domestic Consumption

Author's Notes: I apologize for this post coming out one day later than I originally intended, as I had to resolve a last-minute thing that came up. Next week, I will continue on the normal schedule (Wednesday at 9pm EST). 

The following blog post (except for the introduction) was a course deliverable for the Fall 2019 graduate course "Quantitative-Based Methods for Evidence-Based Policies," where constructing econometric models was a key component of class deliverables. In this course deliverable, using domestic economic data provided by the course instructor, I created several regression models to estimate domestic economic consumption.
Quantitative Modeling of Domestic Consumption

Introduction: As every economics 101 class teaches in the first week, one of the main aspects of Gross Domestic Product (GDP) is domestic consumption, or the amount that citizens spend on domestic goods and services (Chappelow, "Consumer Spending"). To show the extent to which domestic consumption is important to both the economy and GDP, much of the ongoing COVID-19 recession can be attributed to a massive decrease in domestic consumption spending due to job layoffs and increased unemployment (Kharas and Triggs). Generally, this framework emphasizes the fact that the virus's global spread led to social distancing and similar measures that shut down a plethora of industries and sectors ("supply shock"), whose workers became unemployed, thus leading to decreased income and wealth to spend on consumption ("demand shock"), thus further reducing demand for services and goods (Holden). Economists have speculated that economic recovery from the COVID-19 recession will require a complete reopening of the economy and society, something that is dependent on a vaccine or treatment available to most citizens (Carlsson-Szlezak et al.). With this topic being relevant, today's post will create a series of econometric models to predict domestic consumption using other economic variables. 

Theory: Every observation in the dataset represents a particular year from 1950 to 2018. The dependent variable, Consumption, represents aggregate US consumption in 2009 real United States Dollars. To form a model to estimate and extrapolate patterns of consumption, I will be using 8 explanatory (or control) variables. Each variable and the corresponding theory linking each to the dependent variable is below:
  • Total disposable income: Represents the average amount of disposable income per individual in 2009 U.S. Dollars. I expect this variable to be one of the most powerful variables in the model because when citizens in an economy have more disposable income, they have more money to spend on goods and services, thus a positive association should be expected (Cautero). 
  • Year: A quantitative variable that will be used to control for time-based trends in the ata and allow for extrapolation based on past trends
  • Disposable income as savings: The average percent of consumable income out of total income held in savings accounts. Like with total disposable income above, this variable could be a critical variable because as consumers save more money, they have less to spend on consumption. 
  • Civilian unemployment rate: Representing the proportion of Americans in the labor force who are not employed and are actively seeking a job, this variable is also expressed as a %. An increase in the employment rate should be associated with a decrease in consumer spending because as people become employed, they have less income to spend on consumption. 
  • Real public consumption and investment: Represents the total amount of government purchases of goods and services, expressed in 2009 real U.S. Dollars. An increase in government consumption and investment should be associated with an increase in private consumption because government purchases (including public employee salaries) of goods and services pass public money into the hands of consumers (Kenton). 
  • Average prime rate: Expressed as a percent, this variable represents the average prime rate charged by banks. An increase in the prime rate should be associated with a decrease in consumer spending as consumers would be able to earn more by putting disposable income into interest-bearing accounts rather than spending such money (Canady). 
  • M1 money supply: Represents the amount of M1 money in the economy, comprising physical money, demand deposits, and other very liquid forms of currency (Chappelow). An increase in M1 should be associated with an increase in consumption because M1 currency is very liquid and can be easily converted into physical money (if not already), thus allowing consumers to easily spend it (Chappelow, "M1"). 
  • Time deposits: Represents the difference between M2 and M1 money supply, which was self-calculated using the existing data. The variable is supposed to represent the amount of money held in time deposit accounts, as M2 covers less-liquid time deposits (such as savings accounts and mutual funds) in addition to M1. An increase of money in time deposit accounts should be associated with a decrease in consumer spending because if more disposable income is put into time money accounts, less money goes towards consumption as such funds are less liquid. 
Research Design: The raw economic data used for the original course deliverable was provided by the course instructor (Bhargava). In an email correspondence prior to the publication of this post, they were unsure about the exact original source of the data. However, they speculated that it may have come from the World Bank's World Development Indicators database, and noted that such American macro-economic data is readily available from other sources. I was not able to search other sources to find similar or matching data before the publication of this post. 

For each regression model, I will utilize regular OLS regression with robust standard errors to account for possible heteroscedasticity. The usage of the real total disposable income and government consumption are used to account for differences in price changes, because considering the amount of time the data used in this study comes from, there has been a massive increase in nominal prices since 1950 ("Consumer Price Index, 1913-"). Using nominal prices would create interpretation problems because prices have increased drastically throughout this paper’s time period, making the modern US dollar worth far less than what it was worth then ("Consumer Price Index, 1913-"). Using real dollar amounts allows all prices throughout the dataset to be fixed to a single date, allowing for easier interpretation of the data relative to price levels. 

Model 2 will convert several variables (Consumption, Total disposable income, Real public consumption and investment, M1 money supply, Time deposits) to natural logs to account for wide variation in the numerical results in several variables. The other variables (Disposable income as savings (%), Civilian unemployment (%), and Average prime rate) will be kept as normal variables. The four models are constructed as follows: 
  • Model 1a: Yi Consumption = β0 + β1 Year + ε
  • Model 1b: Yi Consumption = β0 + β1 Year X1+ β2 Total disposable income X2 + ε
  • Model 1c: Yi Consumption = β0 + βn [All Explanatory Variables] Xn + ε
  • Model 2: Yi lnConsumption = β0 + lnβn [Natural Logged Variables] Xn + βn [Regular Explanatory Variables] Xn + ε
Results: 
Table 1: Regression Results
*p < .05, ** p < 0.01, *** p < 0.001
Notes: Values in parentheses indicate robust standard errors, which account for possible heteroscedasticity in the data. Variables with a “^” indicate that they were converted into natural logs in Model 2. 

As can be seen in Table 1, in Models 1a, 1b, and 1c, Total disposable income has a very powerful positive association with consumption as predicted, being statistically-significant at the .1% level or greater. After including the other explanatory variables, all except for Real public consumption are statistically-significant at the 5-percent level, while M1 money supply and Disposable income as savings (%) are statistically-significant at the .1% level or higher. Converting several of the explanatory variables into natural logs in Model 2 has three mixed effects. First, the non-natural logged variables of Year and Civilian unemployment rate no longer become statistically-significant. Second, the natural-logged Real public consumption and investment becomes statistically significant at the 5-percent level. Third, the natural-logged versions of M1 Money Supply and Time deposits also become statistically insignificant. At the same time, the logged versions of Total disposable income and Disposable income as savings (%) remain statistically-significant at the 0.1% level or higher. 

Implications: The results of this model creation have several important implications for understanding which economic variables are most associated with consumption. First, as predicted, consumer disposable income is a very powerful indicator, which is further reinforced by the regular results for M1 money supply and the amount in time deposit accounts. Second, government consumption and investment does not have a strong association with consumption, and even turns negative with natural-logging that variable. Finally, the raw civilian unemployment rate has little association with consumption, which may be because in times of high unemployment, by basic economic logic, consumers will still have to consume a certain amount, but will also consume less than normal due to their lower level of income. 

Works Cited:

Canady, Tisa Silver. “How Interest Rate Cuts Affect Consumers,” Investopedia, 31 Jul. 2019, www.investopedia.com/articles/economics/08/interest-rate-affecting-consumers.asp. Accessed 13 Oct. 2019. 

Carlsson-Szlezak, Martin Reeves, and Paul Swartz. "Understanding the Economic Shock of Coronavirus." Harvard Business Review, 27 Mar. 2020, hbr.org/2020/03/understanding-the-economic-shock-of-coronavirus. Accessed 30 Jun. 2020. 

Cautero, Rachel Morgan, “What is Considered Disposable Income,” The Balance Careers, 3 Jun. 2019, www.thebalance.com/what-is-disposable-income-4156858. Accessed 15 Oct. 2019. 

Chappelow, Jim. "Consumer Spending Definition." Investopedia, 12 Jul. 2019, www.investopedia.com/terms/c/consumer-spending.asp#:~:text=Consumer%20spending%20is%20all%20spending,critical%20concept%20in%20economic%20theory. Accessed 30 Jun. 2020. 

---. “M1,” Investopedia, 8 Oct. 2019, www.investopedia.com/terms/m/m1.asp. Accessed 13 Oct. 2019. 

“Consumer Price Index, 1913-,” CPI Calculator Information, Federal Reserve Bank of Minneapolis, www.minneapolisfed.org/community/financial-and-economic-education/cpi-calculator-information/consumer-price-index-and-inflation-rates-1913. Accessed 14 Oct. 2019. 

Holden, Richard. "Vital Signs: COVID-19 recession is different – and we need more stimulus to deal with it." The Conversation, 18 Jun. 2020, theconversation.com/vital-signs-covid-19-recession-is-different-and-we-need-more-stimulus-to-deal-with-it-141037. Accessed 30 Jun. 2020. 

Kenton, Wil. “Government Purchases,” Investopedia, 15 Oct. 2019, www.investopedia.com/terms/g/governmentpurchases.asp. Accessed 15 Oct. 2019. 

Kharas, Homi, and Adam Triggs. "The triple economic shock of COVID-19 and priorities for an emergency G-20 leaders meeting." Brookings Institution, 17 Mar. 2020, www.brookings.edu/blog/future-development/2020/03/17/the-triple-economic-shock-of-covid-19-and-priorities-for-an-emergency-g-20-leaders-meeting/#cancel. Accessed 30 Jun. 2020. 

Data Source:

Bhargava, Alok. In the author's possession. 
     Nathan Parmeter
     Author and Host, The Parmeter Politics and Policy Record

No comments:

Post a Comment