US Consumer Condence Responses to Shocks: COVID19 Scenarios and Recovery. Under review: REHO Special Issue

Personal consumption expenditure has long been a key driver of the US economy, accounting for roughly two-thirds of GDP. Consumer condence is a signicant predictor of consumer expenditure especially in times of shocks. The US has experienced many shocks since 1978; the latest of which is the outbreak of coronavirus. COVID19 has spread from China to the whole world; the pandemic has had far-reaching consequences for global political, social, economic, and nancial structures. Consumer condence is one of the leading economic indicators that provides information on the current and future path of the economy, helps in stimulating economic activity, and predicts change in macroeconomic variables, especially during times of economic and political uncertainty. Our study investigates the relationship between consumption expenditure and consumer condence in the USA. It analyzes the US consumer condence response to 11 different shocks since 1978, focusing on COVID-19 shock. Our investigation uses Michigan's monthly Consumer Sentiment Index (CSI) and its ve components from January 1978 to April 2020. The paper is unique in quantifying the potential variations in US consumer condence due to COVID-19 under different scenarios; by providing a projection until August 2021. The goal is to estimate the time needed for recovery and provide guidance to policymakers on ways to restore consumer condence to tame the impact of coronavirus on effective demand. Under the two more optimistic scenarios we predict that recovery will begin by January 2021. Under the third, less optimistic, scenario we predict that recovery will begin by April 2021. period. or business cycles include peak (local maxima), recession (inection point), trough (local minima), recovery (another inection point), and again peak. We analyze these by building a scatter graph of each period of unique event and adding trend line to see the past trends of peak, recession, trough, recovery, and again peak; then we will be able to form accurate mathematical model to regress the forecasting of the effect of COVID–19 event. In the following we analyze the effect of ve of the unique that affected the USA economy.


Introduction
It is a widely held belief among social scientists that a su cient level of con dence is crucial for stabilizing and maintaining the social, political and economic system (Roth, 2009). Subjective assessments of an economy's recent directions, and perceptions of its possible future prospects, known as consumer sentiment or consumer con dence, have become a key ingredient in predicting the future of the economy (Kellstedt et al., 2013). A Consumer con dence index measures consumers' con dence based on their degree of optimism concerning the economy, this produces an indicator that is designed to capture changes in economic activity and broadly used in macroeconomic forecasts (Karagöz & Altunay, 2015).
Despite ongoing academic debate concerning its ability in forecasting, consumer con dence is broadly accepted to be a leading economic indicator that provides information on the current and future path of an economy, helps in stimulating economic activity, and predicts change in macroeconomic variables (Celik, 2010) especially during times of economic and political uncertainty. The New York Times reported "whatever the shortcomings of the consumer con dence indexes, nearly all the researchers agree that when combined with other data, they provide some additional information in forecasting consumption" (Uchitelle, 2002). Consequently, many developed countries have established consumer con dence indexes to measure and disseminate the latest consumer attitudes (Celik, 2010). Oftentimes, changes in consumer con dence are characterized as "surprising" or "unanticipated" (Kellstedt et al (2013). Periods of high economic or political uncertainty are usually associated with high uctuations in consumer con dence, which consequently affects consumption. Unfortunately, households' willingness to consume/buy is negatively affected by uncertainty (Acemoglu and Scott 1994). Thus, even if the consumers' nancial position is unchanged, the negative effect of higher uncertainty on marginal propensity to consume, can lead a drop-in consumption (Gosselin and Desroches, 2002).
The United States has confronted successive crises and major events since 1978, which broadly affected the country's economic and political performance and deteriorated consumer con dence. The most signi cant of these crises are detailed in chronological order below. Early in 1980, the American economy experienced a recession, regarded at that time as the most signi cant since the Great Depression that began in 1929, caused by the disin ationary monetary policy adopted by the Federal Reserve and the 1979 Iranian Revolution. The recession led to a fall in President Ronald Reagan's approval ratings (Slaying the Dragon of Debt, 2020). Another shock came in August 1991 when Iraq invaded Kuwait in order to gain more control over the lucrative oil supply of the Middle East. The war signi cantly impacted global oil production and caused a sharply increase in oil prices for American households (Garner, 2002). According to Throop (1992) the decline in consumer con dence associated with the war helped in predicting the subsequent decline in consumer expenditure. The 1990s also saw the development of the "tech bubble" or "dot-com bubble", which is considered one of the major economic upheavals in the American equity markets. The bubble was characterized by great upraise in the stock prices of internetbased companies during the period 1997-2000. Only one of every two internet-based companies were able to combat the fall that came with the bursting of the bubble (New York Times, 2000).
On September 11, 2001, nineteen men hijacked four fuel-loaded US commercial airplanes and crashed them into the World Trade Center in New York and the Pentagon in Washington DC. Approximately 2,977 people were killed in these terrorist attacks which also included the crashing of another plane outside of Shanksville, Pennsylvania. (CNN, 2019). The September 11 attacks caused immense losses in both human and economic terms. Seven days later, letters laced with anthrax began appearing in the U.S. mail, followed by the killing of ve Americans and the wounding of 17 in what became the worst biological attacks in U.S. history. In response to the September 11 attacks, the USA invaded Afghanistan and toppled the Taliban that had been in power since 1996 (Witte, 2014). In 2003, the USA initiated war on Iraq, to destroy its alleged weapons of mass destruction and end Saddam Hussein's dictatorial rule, resulting in the deaths of more than 4,700 Americans and more than 100,000 Iraqis (CFR, 2020). The subprime mortgage crisis, which lasted from 2007 to 2010, resulted from the expansion of high-risk mortgages, combined with rising house prices (Duca, 2013). This turmoil was followed by the worst nancial crisis since the Great Depression, which had far-reaching negative consequences for the American and global economy (Ellis, 2009). After that, a novel in uenza virus (H1N1) sprung in 2009. The virus was rst detected in the United States and quickly spread throughout the country and the world. By August 10, 2010, the World Health Organization (WHO) announced the end of the pandemic, which had resulted in more than 12, 000 deaths in the USA (CDC, 2019).
The latest and current shock is the COVID-19 pandemic. The virus emerged in Wuhan, China in mid-December 2019 and still persists globally. COVID-19 was rst reported in mid-December 2019 in Wuhan, China. However, research investigating the pandemic's impact on the economy is still emerging. Research to date includes McKibbin and Fernando's (2020) study, published in February, that explored seven different scenarios on how COVID-19 might evolve in the coming year using Dynamic Stochastic General Equilibrium (DSGE) and Computable General Equilibrium (CGE) modelling. The study examined the impacts of the different scenarios on macroeconomic outcomes and nancial markets in a global hybrid DSGE/CGE general equilibrium model. It showed that under all scenarios GDP growth will reduce across economies globally, and predicted that the development of COVID-19 into a global pandemic will cause the damage done, in terms of lost economic output worldwide, to increase rapidly (reaching in excess of $9 trillion dollars under their most pessimistic scenario). Using a macroeconomic model, Fornaro and Wolf (2020) showed that the COVID-19 outbreak might lead to a demand-driven downturn, followed by a supply-demand doom loop and potential stagnation traps brought about by pessimistic animal spirits.
Similarly, Baldwin and di Mauro (2020) claimed that COVID-19 is both a demand and a supply shock which is likely to dramatically and signi cantly slow down aggregate trade ows. They also argue that a manufacturing distress and supply-side infection is inevitable due to distortions of international supply chains. Wren-Lewis (2020) drew upon modelling of the economic effects of the in uenza pandemic to argue that reduction in economic growth attributable to COVID-19 will result from higher production costs, reduced labor supply, higher temporary in ation, and reduced social consumption. To investigate the pandemic's likely macroeconomic shocks, Barua (2020) utilized a standard macroeconomic AD-AS model to understand COVID-19's impact on economic areas or activities including supply, demand, supply chains, trade, investment, price levels, exchange rates, nancial stability and risk, economic growth, and international cooperation. The study identi ed features that need to be considered when governments and international institutions are designing policy responses to mitigate economic shocks.
These include ensuring that responses are comprehensive, innovative, co-ordinated, and provide extra support for developing economies (including debt reduction).
Furthermore, the pandemic has had far-reaching consequences for global political, social, economic, and nancial structures. By mid-May 2020 the number of con rmed cases reached 4,434,653 and the number of deaths hit 302,169. The pandemic has provoked a dramatic shift in consumer con dence and behaviors (WHO, 2020). The erosion of consumer con dence will make trust more important than ever before, and this will necessitate effective policies targeting con dence-building through different channels. Obviously, the recovery from COVID-19 will happen when consumers regain su cient con dence to signi cantly increase their effective demand.
A growing body of literature has set out to analyze the relationship between consumer con dence (CC) and several economic variables. Most of previous research focused on the predictive power of consumer con dence indexes, with little attention to the in uence of shocks and unique events on consumers' attitudes and con dence, which will impact on their expenditure. Our study lls this gap in the literature by conducting two investigations on the US economy. The rst focuses on the relationship between consumer expenditure and the University of Michigan's Consumer Sentiment Index (MCSI). While the second looks at the effects of speci c shocks on the MCSI. This paper is unique in quantifying the potential variations in US consumer con dence due to COVID-19 under different possible scenarios; by providing a projection until August 2021.
Our goal is to provide guidance to policy makers on ways to rebuild consumer con dence during and after the virus to tame its impact on effective demand/consumption levels. The paper builds upon the experience of US consumer con dence behavior in times of shock since 1978. We use time series data from the MCSI, covering the period 1978:1 to 2020:04, and its ve components to analyze the effect of 14 shocks on the MCSI.
The paper rst summarizes the consumer con dence measures. Sections 3 and 4 discuss the relationship between consumer con dence, economic forecasting and business cycles. The data and methods are presented in Section 5. Section 6 discusses the results of the model analysis, and the ve scenarios simulated post-COVID-19. Section 7 concludes the paper's main ndings and discusses some policy implications.

Consumer Con dence Measures
Consumer con dence is a major indicator for analysts and policymakers, especially in times of disturbances, since consumption contributes approximately two-thirds of real GDP (Fuhrer, 1993).
There are two approaches to looking at the role of con dence. The rst is the "Animal Spirit" which poses autonomous values variations that affect economic activity. This approach considers the psychological factors that can in uence consumers' decisions (Gosselin and Desroches, 2002), as exogenous variables. The second approach is the "news" or "information" which deals with con dence as an endogenous variable, and a re ection of current macroeconomic conditions. This view suggests a connection between consumer con dence development and subsequent macroeconomic activity (Lachowska, 2013). Barsky and Sims (2012) found that con dence is a re ection of news that provides essential information about current and future economic situations. Likewise, Cochrane (1994) reported that consumption shocks are proxies for news that consumers are receiving about future productivity that does not otherwise appear in the information sets of econometricians. While Hall (1993) and Blanchard (1993) reported that the exogenous movements in consumption were the cause of the US recession in 1990-91. We believe that consumer con dence is formed from a blend of two sources psychological factors and information about macroeconomic conditions, where the latter heavily affects the former. Consumer con dence re ects speci c attitudes related to particular events and the economic situation. Consumer spending is affected by each consumer's con dence as well as their current income and wealth. Both willingness to buy and affordability creates the consumer's effective demand. Willingness to buy is partially derived from consumer con dence.
There are two widely followed measures of consumer con dence in the United States. The Consumer Sentiment Index issued by the University of Michigan (MCSI) and the Consumer Con dence Index (CCI) published by the Conference Board. Both indices are based on responses to ve survey questions; two questions ask respondents to assess their present economic conditions, which receives 40% weight of the index, and the other three tackle consumers' expectations (Dion, 2006). Our current study uses the MCSI.
The MCSI started annually in the 1940s as the rst US survey to measure, understand, and analyze the impact of changes in consumer attitudes and expectations (Dion, 2006). The MCSI became a quarterly index in the 1950s and has been available on a monthly basis since 1978 (Howrey, 2001). The index contains 50 core questions covering different aspects of consumer attitudes and expectations. The survey polls a sample of 500 people by telephone and asks questions focusing on their present and future nancial conditions, spending intent, and business conditions (Michigan University, 2020). The MCSI re ects recent changes in the economy rather than the level of economic activity (Bram and Ludvigson, 1998). A higher value of the MCSI indicates greater optimism among private households.

Consumer Con dence And Consumer Expenditure
Consumer con dence is a leading indicator of consumption and thereby the MCSI helps to forecast changes in consumption expenditure independent from other indicators (Juster and Watchel, 1972;Garner, 1981).
The idea of using consumer con dence for consumption prediction goes back to 1963 (Croushore, 2004), when Eva Mueller (1966) found that consumer con dence is a signi cant explanatory variable for consumption spending in a 10-year regression with lagged consumption. Carroll et al. (1994) claimed that lags of MCSI in the US have explanatory power for household spending changes. Increased uncertainty drives households to cut their spending, while habit formation delays the adjustment of consumption spending. In the same vein, Howrey (2001) reported the usefulness of the high-frequency MCSI information, since the monthly MCSI information helped improve quarterly forecasts. Delorme et al. (2001) and Brown (1997) concluded that con dence measures can better predict consumption of durable goods compared to non-durable goods. The MCSI has a signi cant forecasting power explaining consumption of automobiles (Juster and Wachtel, 1972). Bram and Ludvigson (1998) reported a signi cant incremental predictive power of the MCSI for forecasting consumption growth, with some questions having more predictive power than others. In the same vein, Wilcox (2007) showed that MCSI sub-indices signi cantly improve the consumption forecasting compared to the aggregated index. Carroll et al (1994) have also reported the predictive power of the MCSI lags for future consumption changes. Howrey (2001) concluded that the MCSI improves the forecasting accuracy of recession, but it produces only little improvement in forecast accuracy for consumption expenditure.
Over the past decades, personal consumption expenditure has been a key driver of the US economy, especially during the times preceding a recession (Emmons, 2012). It accounts for the largest share of GDP in the economy, roughly two-thirds (Toossi, 2002). In 2019, when consumer con dence hit a 20-year high, consumer spending accounted for roughly 80 percent of real GDP growth (Council of Economic Advisers, 2020). Furthermore, consumer consumption determines where employment is generated, as consumer choices determine whether employment is in the nal-goods industries or in related intermediate industries, which in turn affects the GDP level (Toossi, 2002).

Consumer Con dence And Business Cycles/shocks
The MCSI summarizes the macro-economic shocks effect, since it contains questions that help forecasting or explaining consumption through their in uence on it, the index also leads future consumer decisions (Dion, 2006). The MCSI con dence measure has a major role in understanding business cycle uctuations (Carroll et al, 1994;Ludvigson, 2004;Benhabib et al., 2015). It summarizes the macroeconomic shocks effect, since it contains questions that help forecast consumption through their in uence on it, and on future consumer decisions (Dion, 2006). Going back to 1936, Keynes in his macroeconomics theory argued that "waves of optimism and pessimism" could be major drivers of business cycles. Among others, Taylor and McNabb (2007) demonstrated the pro-cyclicality of consumer con dence and its signi cant role in predicting downturns.
According to Christiansen et al. (2014) consumer sentiment holds greater predictive power for US recessions than both the classical recession predictors and the common factors. Additionally, Throop (1992) argued that although con dence usually re ects the current economic situation, it can predict independently the direction of consumer spending at times of unique events, so sentiment provides useful information about future consumer expenditures in uncertain times that is not otherwise available.
Nevertheless, consumer sentiment is reported to be more predictive with high consumer con dence volatility during uncertain times (shocks) compared to normal times (Desroches and Gosselin, 2002). The longevity of both the Great Depression and the 2007-08 nancial crisis was because of consumer con dence collapse (Dees & Brinca, 2013). According to Matsusaka and Sbordone (1995), unpredictable changes in consumer sentiment, "Granger-cause", cause changes in real GNP, ranging between 13 and 26% of its total variance. Matsusaka and Sbordone (1995) ascribe their results to uncertainty. In line with this fact, Santero and Westerlund (1996) concluded that strong variations in con dence, which are likely driven by major events, are often followed by uctuations in GDP. Consequently, we can say that there is some evidence that sentiment measures unexplained by economic fundamentals are associated with spending shocks (Oh and Waldman, 1990). Nofsinger (2012) demonstrated the household behavior in boom and bust economic cycles with a particular focus on the 2007-08 nancial crisis. He reported more consumption and less savings in boom times and the opposite in bust time which drags down an already sinking economy.
Dees (2017) used survey data on consumer sentiment to identify the causal effects of con dence shocks on real economic activity in a group of advanced economies. He found that shocks have a signi cant effect on consumption and real GDP, where con dence shocks explain a large variations in real economic activity and hence are partially responsible for business cycle uctuations.

Research Design and Data
The study uses US monthly MCSI3 Index data from January 1978 until April 2020, to investigate the effect of unique events/shocks on consumer con dence. We obtained the MCSI data from surveys of consumers performed by the Survey Research Center of Michigan University. The Index value is affected by three main consumer perception factors; Personal Finances (X1and X2), Economic Conditions (X3 and X4) and Household Goods Buying Conditions (X5). The considered events/shocks start with the 'Early 1980s Recession" and ends with COVID-19, the list of shocks is presented in Table (5) in the appendix. We use Dummy Variable Dum = 1 during the unique event and Dum = 0 otherwise.

Methods
The study conducts a number of residual diagnostic tests for the data set. Serial Correlation Test; Correlogram Squarred Residuals, Correlogram Q-statistics, ARCH LM Test, and Histogram Normality Test.
The study uses Ordinary Linear Regression (OLS) to test the relationship between consumer expenditure and MCSI. OLS is further used to regress MCSI on its components and the shock dummy variable. MCSI= X 1 +X 2 +X 3 +X 4 +X 5 +Dum Equation (1) X1 We are interested in how people are getting along nancially these days. Would you say that you (and your family living there) are better off or worse off nancially than you were a year ago?"

X2
Now looking ahead--do you think that a year from now you (and your family living there) will be better off nancially, or worse off, or just about the same as now?"

X3
Now turning to business conditions in the country as a whole--do you think that during the next twelve months we'll have good times nancially, or bad times, or what?"

X4
Looking ahead, which would you say is more likely--that in the country as a whole we'll have continuous good times during the next ve years or so, or that we will have periods of widespread unemployment or depression, or what?"

X5
About the big things people buy for their homes--such as furniture, a refrigerator, stove, television, and things like that. Generally speaking, do you think now is a good or bad time for people to buy major household items?
Dum Dum=1 during the unique event and Dum= 0 otherwise Further, we build ve regression models to analyze the effects of unique events on the US MCSI. Before we apply the model for prediction, we build three scenarios based on the percentage change in the trend of independent variables.

Results
Consumer Expenditure and Consumer Con dence The relationship between US consumer expenditure and consumer con dence (1978-2020) is presented in Fig (1). The results in Table (1-a) in the appendix show a strong relation between consumption and the MCSI and its lagged value with R-squared equals 0.999. The elasticity of consumption with respect to the MCSI is 22%. The residual diagonistic tests in Tables (1-b, 1-c, and 1-d) in the appendix con rms the absence of serial correlation and heteroskedasticity.

Consumer Con dence Model
The time series trend of US consumer con dence from 1978 to 2020 in gure (2) indicates ARCH effect in the data. Because we can notice that periods of low votality are followed by periods of low volatility, and periods of high volatility are followed by periods of high volatility for prolonged periods.
The results of Equation (1) Tables (4 a, b, c, d, e) in the appendix, indicate a sound impact of unique events on each of the ve components of MCSI (X 1 , X 2 , X 3 , X 4 , X 5 ). However, the MCSI components sensitivity to dummy variables shows great variations, X 3 seems to be the most sensitive/vulnerable component for the shock/unique events followed by X 5 , whilst X 4 appears to be the MCSI component that is least sensitive to the shocks.

Scenario Analysis
We observed that percentage change in the trend of independent variables is between 5% to 20% decrease or increase. Accordingly, we built ve scenarios at 2.5%, 5%, 10%, 15%, and 20% decrease / increase of the independent variables. We estimate MCSI values up to December 2020, this is valid only under the assumption that the present situation persists, and no economic/business cycles resilience occur. After that we regress the trend line for MCSI and its curvature to predict data until August 2021.
The point to be noted is that the business cycles and other economic conditions also have effects on consumer con dence (MCSI) curves. There would also be the combined effect of the COVID-19 event and business cycle activity; if the business e ciency improves quickly MCSI might be on maxima irrespective of COVID-19 or there might be a combined effect of COVID-19 and business cycles on minima and maxima of the MCSI curves; so, we have to apply both models -scenario analysis from 2.5% to 20% values and application of mathematical model for forecasting the MCSI.      The effects of short-term business cycles on consumer con dence caused by unique events Our data show a clear short-term and long-term volatility. We calculated the magnitude and curvature of business cycles and economic activities during the period of each unique event. That illustrates times when the economy recedes and reaches a trough and other times when the economy takes a recovery path and reaches a peak. For the prediction of the effect of COVID-19 on consumer con dence, we start in this section by analyzing the situation around the unique events and economic recovery period.
Economic or business cycles include peak (local maxima), recession (in ection point), trough (local minima), recovery (another in ection point), and again peak. We analyze these by building a scatter graph of each period of unique event and adding trend line to see the past trends of peak, recession, trough, recovery, and again peak; then we will be able to form accurate mathematical model to regress the forecasting of the effect of COVID-19 event. In the following we analyze the effect of ve of the unique events that affected the USA economy.

Unique Event No.1: Early 1980s recession
Though the event lasted from July 1981 to November 1982, its effect was evident from Aug-Sep 1982. From May-81 to Aug-81 consumer con dence followed its own business cycle and was not affected by the unique event. Referring to gure (3), from 0 to 1 refers to May-81 (the whole month), while from 1 to 2 refers to June-81, by the same way 15 refers to Figure (3) US MCSI -Early 1980s recessionFigure (3) US MCSI -Early 1980s recessionJuly-82 which ranges from 14 to 15and so on.
The curvature of the MCSI was at its lowest from Nov-1981 to Aug-1982; after 25 months it was again on its peak at May-1983. Figure (3  We use the same function to forecast the MCSI in 2021 (Table 6) below.The results showed the rst local maxima (peak) during the month of Nov 2019 until Feb 2020. The recessionary slope started in March 2020 and will continue until Dec 2020 (trough). Jan 2021 will be the starting month for a recovery path, until reaching the peak in the months of June, July, and August 2021.  In the real world both business cycles and other economic conditions have effects on consumer con dence (MCSI) curves. The COVID-19 effect would be combined with the effect of business cycle activity; if the business e ciency improves quickly MCSI might be on maxima irrespective of COVID-19 or there might be a combined effect of COVID-19 and business cycles on the minima and maxima of the MCSI curves. Accordingly, we had to apply both models, scenario analysis values in addition to the mathematical model for forecasting. We calculated the magnitude and curvature of business cycles around the ve shock periods considered in our study. We were able to identify the mathematical models of short run business cycles, which show the peak, trough, and the recovery path after each shock. The rst shock period "the early 1980s recession", did not affect the economic activity though it follows its own business cycle activities, the MCSI curve took its recovery path in September 1982. However, the recession effect lasted until November 1982. It took 25 months for the MCSI to reach its peak again. The second shock period "Gulf War" lasted from August 1990 to February 1991. The trough is observed from October 1990 to December 1990; and the MCSI took a recovery path from January 1991 to July 1991, the index reached its peak during March 1991 to July 1991. In the third shock period, we considered four unique events which happened from 2000 to 2001, the events caused a recessionary slope from April to September 2000. The MCSI reached its trough by September 2001 while it took a recovery path from November to reach its peak in January 2002, just two months after the events. Whilst in the fourth shock The study proceeded with the forecast of the impact of COVID-19 on consumer con dence until August 2021, in order to predict the time needed for recovery. We regressed the trend line for MCSI and its curvature under three scenarios. Under the most two optimistic scenarios (2.5% and 5%) the recovery is expected to start by January 2021. Under the less optimistic scenario (10%), we expect the recovery to start by April 2021. The rst two scenarios expect the recessionary trend to continue from March 2020 until December 2020 (trough), the recovery starts from January 2021 and reaches its peak in the months of June, July, and August 2021. While, the third scenario expect the recessionary trend to continue from March 2020 until March 2021, the recovery path starts by April 2021, and the local maxima (peak) is expected in the months of August and September 2021.
[i] From MCSI time series, most of the minimum and maximum values fall under absolute values of 2.5% to 20% values. So, we built our scenarios from 2.5% to 20% expecting that no exaggeration might be caused.

Declarations
The Work Under Consideration for Publication Did you or your institution at any time receive payment or services from a third party (government, commercial, private foundation, etc.) for any aspect of the submitted work (including but not limited to grants, data monitoring board, study design, manuscript preparation, statistical analysis, etc.)? : The authors declare no received payment or services from a third party  Forecasting MCSI-Scenario 02 at 5%

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. Appendix.docx