Masters Thesis

Analysis of consumer sentiments ability to forecast the service industryBy: Steven Kennedy

July 2, 2015

Introduction

The role of consumer confidence, as an economic indicator, has long been studied by many economists’ dating back to John Maynard Keynes. First referred to as “animal spirits” by Keynes it is believed and proven through many empirical studies that there is a relationship between consumer confidence1 and the real economy. More specifically, how a household feels about their present and future economic situation today will have an effect on how they spend their money tomorrow. That being said there is useful information, independent of other economic indicators, that is contained in these consumer confidence surveys. This intuitively makes sense because this survey is the only such economic variable that provides insight about why the consumer made their decision. Therefore, confidence surveys may contain useful information in terms of forecasting future values of certain economic variables such as consumption.

The economy is measured by real GDP, which is made up of four main components: consumption, investment, government spending, and net exports/imports. The largest contributor of these components is consumption, which encapsulates nearly 70% of total real GDP. In 2014, $11.929 trillion of the total $17.418 trillion produced in the U.S was from consumption. Additionally, consumption is broken down into three sub categories: durable goods, non-durable goods and services. For example, a durable good would be something such as a car or TV. A non-durable good would be food or fuel and a service would be like getting a haircut. The three of these sub-components capture all of the personal consumption expenditure in the economy.

The services industry is the largest component of the three but this was not always the case. The services industry has grown over time and contributes much more to the economy then it did back in the 1960’s. As a country evolves from a low-income country to a high-income country the relative layout of its three sectors changes, those being agriculture, industry and services. More specifically, as a country becomes more developed and its income per capita rises it will tend to move its main production from the agricultural sector to the industrial sector and then finally to the services sector. As other less prominent countries can produce the lower level goods and trade them. These shifts are called industrialization and post-industrialization. As the economy goes through these structural changes, the peoples demand for food reaches its natural limit, and therefore leads to the increase in demand for more industrial goods. Once an economy grows even larger, people begin to demand less material goods and more services such as education, health and entertainment. The catch here is that labor productivity in the services sector is more expensive because the service industry relies more upon human capital than natural capital. Human capital is more expensive because it takes time, effort and money to train individuals in the particular service they will provide. This

1 Throughout the paper, the word confidence and sentiment are used interchangeably.

makes services more expensive relative to agricultural and industrial goods and therefore increasing the share of services in the total real GDP.

That all being said it is obvious that the U.S economy is in its post-industrialization phase because it is one of the most developed country’s in the world and consequently the services industry is a main component of the economy’s overall output. For the U.S in 2014, $7.96 trillion in services was produced, which was 45.7% of GDP that year. Since the service industry has now become such a large part of the U.S economy it would behoove anyone to try and understand the economic agents that affect it. Papers have explored this avenue in terms of consumer confidence and have proven that consumer confidence does in fact have useful information for forecasting consumption and more specifically the service industry2. Recent literature stops here and fails to dig deeper into the service industry. In this paper the service industry will be broken down into seven sub-components, as categorized by the Bureau of Economic Analysis, to test which sectors of the service industry are most effected by consumer confidence. Furthermore, a time aspect will also be tested in that I will see how consumer confidence’s predictive power over the service industry has changed over time. In the following section the data and the methods used will be discussed.

Data and Methods

Consumer confidence is currently measured predominantly by The University of Michigan's Consumer Sentiment Index and the Conference Board's Consumer Confidence Index. Both of these indexes turn qualitative data into quantitative data, so that a precise number is given for how confident consumers are. These index’s both implement survey tactics to gather their data, and then they turn these survey responses into a number. Each index asks consumers about their present and future feelings about the economy and their overall economic well-being. The methodology for constructing each respective index is different. Sample size, survey questions and overall computation differ for both3. For this particular analysis the University of Michigan's Consumer Sentiment Index is used because this data is readily available to the public for free. Furthermore, the University of Michigan constructs an overall index and an expectations index. The expectations index is made up of the survey questions that ask consumers about what they think of future conditions of the economy and their overall economic well being. The overall index encapsulates the expectations component and present conditions component. For this analysis, only the overall index was used. The frequency of sentiment data is usually monthly, but because the service variables are recorded quarterly, quarterly sentiment data was also used to keep the same frequency throughout variables.

The second set of variables used is the individual sectors of the service industry. The BEA breaks down the service industry into several sub components.

2 Papers that link consumer confidence to consumption: Ludvigston(2004), Lahiri, Monokroussos, Zhao(2014) and Starr(2012)3 For further interpretation of the consumer confidence index’s consult Curtin(2007)

For this paper seven individual sectors of services were used, along with an eighth service variable that was defined as all services. This eighth variable is used to test if there is significance with the service industry as a whole. The seven sectors are: Housing and Utilities, Health Care, Transportation, Recreation, Food, Financial Services and Insurance, and Other. All of these are also divided into more specific sub-components but for simplicity only the overall sectors are used. All of the service variables are recorded on a quarterly frequency and are seasonally adjusted. Also, the service variables used are percent change from the previous quarter, not just the pure aggregated value for that particular quarter. For further breakdown of these variables see the BEA handbook on Personal Consumption Expenditure table 5.B4.

There are a total of eight different service variables being tested as dependent variables and for each of these four different time periods were tested. There are a total of four regressions per service variable. The point of this was to try and see if the role of consumer confidence has changed as time has progressed and as the service industry has become more prominent in the economy. The four time period’s recursively get smaller in length. The first time period is from 1970 Quarter one until 2014 Quarter four. Then each additional time period removes the last ten years. So, the next time periods are 1980Q1-2014Q4, 1990Q1-2014Q4, and finally 2000Q1-2014Q4.

The purpose of this paper is to test if there is any forecasting ability of consumer sentiment for the individual service industry sectors. So, when running the regressions the variables that were tested for significance were the past values of consumer sentiment. Naturally to see if a variable can predict another, we must use past values of that said variable and observe if it is significant in predicting current values of the consumption variable. Therefore lags of one, two, three and four quarters were used to test if consumer sentiment from previous quarters affects consumption in the current quarter.

Additionally, correlation does not mean causation in any sense. To truly test if consumer sentiment has any sort of causal relationship on the service variables, granger causality methodology was implemented. Meaning, that to accurately assess these relationships one must saturate each individual model with as many significant lags of the dependent variable. This is so that the usefulness of consumer sentiment can be tested after accounting for all past information of the service variable. This can be done by implementing an auto-regressive model and using t-tests’ to see how many lags of the dependent variable to use in each model. Once the amount of lagged dependent variables is determined, we can then start to add our additional lagged consumer confidence variables. This will truly show how effective consumer confidence is at forecasting the individual service variables. With the autoregressive model all past information about that said dependent variable is included and the additional worth of adding lagged consumer confidence can be evaluated. As stated by Diebold, “Lagged dependent variables absorb residual serial correlation and can dramatically enhance forecasting performance5” therefore the

4 Link to BEA handbook: https://www.bea.gov/national/pdf/NIPAhandbookch5.pdf5 Quote from Elements of Forecasting 3 rd edition by: Francis X. Diebold

https://www.bea.gov/national/pdf/NIPAhandbookch5.pdf

use of an auto-regressive model is useful in terms of serial correlation as well. Stationarity is the last concern that must be dealt with before any sort

of forecasting or tests can be run. All variables in question must be either already stationary or made stationary. The consumer sentiment index and the service variables all proved weakly stationary and no differencing was necessary. To test for this, Augmented Dickey Fuller tests were implemented.

Empirical Analysis

The first service variable that will be discussed is overall services; this encapsulates all of the service variables in one. This will at least give some insight to see if consumer sentiment affects the service industry as a whole. Chart 1.1 shows the consumer sentiment and the service industry compared to each other.

Chart 1.1

The service variable line is much more volatile, where as the sentiment line is more smooth. It does appear from simple visual analysis that consumer sentiment and the service industry follow the same patterns for the most part. To further diagnose the relationship, regression analysis was used. Chart 1.2 below shows the statistical results for the four different regressions run for the four individual time periods tested. The p-value’s are reported in parenthesis. First, the one quarter lag of the dependent variable was significant for every time period but then adding a second quarter lag was not. Therefore each model was only saturated with one

auto-regressive term. Then lagged values of consumer sentiment were added to the

Chart 1.2

Note: Any cell block denoted with an ** means the variable proved insignificant. Source: Authors computation

Time Span Intercept 1 Quarter Lag of Dependent Variable

2 Quarter Lag of Dependent Variable

1 Quarter Lag of Consumer Sentiment




Adjusted R-Square

Root MSE

1970Q1-2014Q4

-0.44590(0.5916)

0.43816(<.0001)

** 0.02579(0.0164)

** ** ** 0.2809 1.57729

1980Q1-2014Q4

-1.60092(0.0874)

0.37816(<.0001)

** 0.03921(0.0014)

** ** ** 0.3178 1.51041

1990Q1-2014Q4

-2.82516(0.0037)

0.32339(0.0016)

** 0.05275(<.0001)

** ** ** 0.4407 1.19773

2000Q1-2014Q4

-2.13267(0.0472)

0.07941(<.0001)

** 0.34592(0.0099)

** ** -0.03781(0.0081)

0.5380 0.88222

regressions. For every time period it is apparent that consumer sentiment does have forecasting ability for at least one quarter prior to the horizon target with p-values that show strong significance. Also, as one would expect the parameter values for confidence are positive meaning that when confidence increases so does consumption. The increasing adjusted r-square and the diminishing root MSE prove that as time has progressed the role of consumer confidence on the service industry has amplified. This could simply be due to the fact that the service industry has grown through the passage of time. As seen in the last row of the chart, a consumer sentiment lag of 4 quarters also proves to be significant. Which means that consumption of services is also affected by how confident consumers were a year ago. This initial test of overall services shows that consumer confidence does play a role in the consumption of services. Furthermore as time has progressed the magnitude at which consumer confidence affects consumption has also increased. Now we must dig deeper into the service industry itself and identify which sub-components are affected most by consumer confidence.

The next service variable that will be discussed is Housing and Utilities. These services are comprised of any sort of housing service provided, such as apartment rental and also any sort of utility needed for that housing, like electric and gas. Chart 2.1 shows the relationship between consumer confidence and housing and utilities. Again like the overall service variable from before the housing and utilities variable is more volatile. This is most likely due to the nature of the data. The service variables are in the form of percent change so the line will look noisier then the sentiment line. Due to this fact it is difficult to assess any sort of relationship between these two variables. A rough initial analysis would suggest that consumer confidence and the service variable follow similar paths. Which is

seen easily from 1970-1985 and during the sharp decrease in early 2000. This is something we would expect because if consumers feel less confident they will purchase either less expensive housing options or decide to clamp down on utilities by using less electric and gas where they can to save money.

Chart 2.1

Chart 2.2Time Span Intercept 1 Quarter

Lag of Dependent Variable





Adjusted R-Square

Root MSE

1970Q1-2014Q4

-0.14258(0.9086)

** 0.03225(0.0285)

** ** ** 0.0212 2.41608

1980Q1-2014Q4

-2.27542(0.0853)

** 0.05203(0.0008)

** ** ** 0.0719 2.27066

1990Q1-2014Q4

-2.95480(0.0443)

** 0.05724(0.0009)

** ** ** 0.0982 2.09175

2000Q1-2014Q4

-4.15524(0.0193)

** 0.06962(0.0013)

** ** ** 0.1503 1.98812

Note: Any cell block denoted with an ** means the variable proved insignificant.

Source: Authors computation

The statistical analysis sheds light on the relationship between these two variables. Chart 2.2 displays the results of the four regressions run. A very interesting thing happened when I tested the lag of the dependent variable. For all time periods it proved that the lagged dependent variable for housing and utilities was insignificant. This is a peculiar find because one would assume that for almost all economic variables, the previous value for said variable would be significant in explaining the present value. This find was concerning and interesting in that for every time period an auto-regressive model could not be used, which directly goes against the method I was using for the tests. So, granger causality methodology cannot be strictly enforced and makes the difference between correlation and causality difficult to distinguish. Nonetheless, I still tested if lagged consumer sentiment was significant in predicting housing and utilities.

As scene in the chart, a one quarter lag of consumer sentiment was significant for every time period. The parameter values are also positive, following what we would expect. Another result from the tests shows that as time has progressed the adjusted r-square increased and the root MSE decreased. This illustrates that the effect of consumer sentiment on housing and utilities has increased over time. This is most likely caused by the increase in the need of housing and utility services as time has gone on. The results show that if one would like to forecast next quarters housing and utilities totals, they could use consumer sentiment from this quarter.

These results make theoretical sense as well because one would expect to try and spend less money on utilities and there housing if they were not feeling confident about their overall economic well being and vice versa. Also, being able to manage ones utility bill is quite easy if you really are determined. In the winter if a consumer isn’t feeling good about their economic well being they can simply shut off their heat and bundle up under a blanket to save money and the same goes for electricity. The other side of this service variable is housing. The BEA defines these services as rental housing to tenants, which is also further broken down. Since these services are rentals, most people pay their rent monthly and are usually not bound by a contract when only renting. Therefore, if a person is feeling economically depressed they can simply move out and find new cheaper housing and vice versa. That being said, it is clear how consumer sentiment can affect a persons housing choice. Naturally, if someone expects their future income to increase they will feel confident and purchase a more expensive apartment.

The next service variable that we will look at is financial services and insurance. This variable is broken down into two components: services that are provided in consulting and managing someone’s financial assets and insurance. Insurance is a service in it of itself because it provides something that is neither durable nor non-durable. It is simply a safety net that a company provides to you if anything should happen. Hence the company is providing a service for you by protecting you in time of need. The financial advisory side is clearly a service because a consumer must pay a person to provide them with knowledge and advice. The consumer is receiving pure guidance and that is what they are paying for along with the commissions for the financial institutions. Chart 3.1 shows the connection between sentiment and this service variable.

Chart 3.1

The financial services and insurance variable is volatile and has large oscillations throughout the series. This is due to the fact that the financial market is very sensitive. That being said it doesn’t appear from strictly a visual stand point that sentiment affects the financial services industry. The statistical analysis in chart 3.2 will provide more clear conclusions.

Chart 3.2Time Span Intercept 1 Quarter

Lag of Dependent Variable






Adjusted R-Square

Root MSE

1970Q1-2014Q4

2.44579(<.0001)

0.41751(<.0001)

** ** ** ** ** 0.1700 5.51082

1980Q1-2014Q4

2.06877(0.0002)

0.48605(<.0001)

** ** ** ** ** 0.2330 5.46083

1990Q1-2014Q4

-4.53633(0.1605)

0.46963(<.0001)

** ** 0.07372(0.0548)

** ** 0.2914 4.48536

2000Q1-2014Q4

0.72673(0.2103)

0.46853(<.0001)

** ** ** ** ** 0.2304 4.20937

Note: ** means the variable was not significant. Computation done by the author.

First off, lets look if an auto-regressive model could be used for this variable. Contrary to housing and utilities, a one quarter lag of the dependent variable proved significant. So, our model is of an auto-regressive nature and the auto-regressive term is strongly significant. Consumer sentiment though was insignificant for every time period and every lag length. There is only one time that consumer sentiment proved significant and it was for the time period 1990-2014 and for a two quarter lag. Since the significance only occurred once throughout all the tests that were run, it is hard to say that consumer sentiment effects the financial services industry. Also, the p-value in that instance is .0548, which doesn’t prove significance at the 5% level.

There are some possible reasons why consumer sentiment wouldn’t affect the financial services and insurance industry. Looking at the insurance side of this service variable, it isn’t that far-fetched that consumer confidence wouldn’t have any affect. In the world we live in today and specifically the United States insurance is for the most part required. Observing a current event, Obamacare is a recent act that forces every American to have health insurance. Consumers don’t have a choice by any means, so consumer confidence cannot play a role in the purchase of at least health insurance. Car insurance is also a requirement if someone purchases a car and homeowners insurance is required if you buy a house. There is no choice involved with these decisions, so yet again consumer confidence cannot play a role. Of course there are other types of insurance that are not required by law, such as life, occupation, and travel insurance. These are obviously wants and not needs, so therefore consumer confidence could be a factor. Nevertheless, insurance is a necessity in the world that we live in today. Because of the unpredictability of the world, everyone regardless of who they are want to feel protected just incase something happens to them. The shear threat of losing something or having to pay an excessive amount for a lose that was not predicted makes insurance a no brainer to consumers.

The other side of this service variable is financial services. The BEA breaks this down into commissions paid to financial institutions for their services, such as portfolio management and investment advice. Most people are not well versed on financial products and don’t have the time to spend learning about them. So, they hire financial experts from accredited companies to manage their assets. Whether or not a consumer is feeling confident, they will surely always want someone managing their assets to make sure that their money is taken care of. Thus, consumer confidence doesn’t really affect the financial services industry because the purchase of this financial service is always necessary.

The next service variable that will be discussed is transportation services. This variable is comprised of motor vehicle repair/maintenance and any sort of transportation service. These transportation services include railway, taxicab, intercity mass transit, and air transport. There is also a water transportation component as well. Chart 4.1 shows a graphical representation of this variable compared to consumer confidence. The transportation variable doesn’t appear to be that volatile, compared to the previously discussed service variables. The peaks and troughs of the transportation variable seem to match consumer confidence quite

well. This can easily be observed by looking at the time period from 1970-1985 and 1995-2010.

Chart 4.1

The increases and decreases in sentiment match the transportation series. This would lead to assume that consumer confidence probably plays a role in the transportation service industry. To further test these claims, chart 4.2 demonstrates the statistical results.








Adjusted R-Square

1970Q1-2014Q4

0.64685(0.0446)

0.56857(<.0001)

0.19085(0.0101)

** ** ** ** ** 0.5068

1980Q1-2014Q4

0.60861(0.0746)

0.56794(<.0001)

0.22783(0.0058)

** ** ** ** ** 0.5844

1990Q1-2014Q4

0.40322(0.2652)

0.61333(<.0001)

0.20398(0.0435)

** ** ** ** ** 0.6028

2000Q1-2014Q4

4.56926(0.0531)

0.78924(<.0001)

** ** ** ** ** -0.05560(0.0482)

0.6063

Chart 4.2

Note: ** means the variable was not significant. Computation done by the author. Again, the auto-regressive terms must be checked first. For this particular

variable, lags of up to two quarters proved significant. Only one lag proved significant for the final time period of 2000-2014. Nonetheless the model used for every time period was of auto-regressive nature. Now that the model is saturated with as much past information as possible, lags of consumer sentiment can be tested. Contrary to the conclusion drawn from chart 4.1, the lags of consumer sentiment don’t prove to be significant. The one exception to this is in the time period of 2000-2014, where a lag of four quarters of consumer sentiment proves to be weakly significant at the 5% level and has some sort of predictive power over transportation services. An interesting observation is that the parameter estimate is negative, meaning that when consumers are more confident they use less transportation services. This slightly makes sense because the transportation variable is comprised of mostly public mass transportation services so when consumers are more confident about their economic well being they may chose to possibly use less public transportation and more private transportation, such as purchasing a car.

The lack of significance of the lagged consumer sentiment variables proves that consumer sentiment doesn’t affect the transportation services industry. This could be because people always need some sort of transportation to get them where they need to be. Transportation is a necessity and so is vehicle maintenance. When your car is not working you have to go fix it, there is no other way around it. Consumer confidence may play a role in the purchase of private transportation, like a car. When someone is going to invest in a durable good, like a car, how the consumer feels about his or her economic situation will definitely shape their decision making. Public transportation is a service, not a good and is always readily available. Despite how a person feels about their economic well being, they will always have the option and need to purchase public transportation if they have to get somewhere. Another aspect of the transportation variable is water transportation. Water of course is a necessity, and therefore water transportation is always going to be essential. So, again consumer confidence will not affect this aspect of the transportation services.

The significance of the four quarter lag of consumer sentiment could mean that sentiment is becoming a slight factor in the transportation service industry. Even though the lag is of a full year, the p-value does prove significance and further analysis will be necessary to see if sentiment has truly become a a good predictor of the transportation service industry.

Looking back at the argument made earlier about how on chart 4.1 the transportation series and the consumer sentiment series follow the same paths. It could be possible that current consumer sentiment and not lagged consumer sentiment effect transportation in the current quarter. Still, the purpose of this study is to see the forecasting ability of consumer sentiment on service industry variables. The only way the use of current sentiment would be useful is to forecast what sentiment would be in the next quarter, so that next quarters transportation services could be forecasted. This is not a reliable practice because using a forecasted variable to forecast another variable would give uncertain results.

Healthcare is the next service variable that will be examined. Healthcare is broken down into paramedic, dental, and physician services. It also is comprised of hospital and nursing home services. Chart 5.1 depicts the relationship between sentiment and the healthcare series.

Chart 5.1

The healthcare series appears very volatile from 1970-1983 but then appears to become less noisy after that. The two series appear to have no correlation to each other at all. They don’t follow similar paths. This would mean that the two variables will more then likely have no connection to each other. Chart 5.2 will further assert these claims.

The auto-regressive terms do prove strongly significant for up to a one quarter lag. So the model used for every time period is saturated with at least one auto-regressive term. Similar to the past few service variables already explored, the lagged consumer sentiment variables prove insignificant. The one exception is for the period of 2000-2014, the two quarter lag of consumer sentiment proves weakly significant at the 5% level. So, as of more recent years, an argument could be made that consumer confidence plays a minor role in the purchasing of health care services but only because of the weak significance of the two quarter lag of consumer confidence. The parameter estimate is positive, so when confidence increases it seems that the demand for health care increases. This intuition makes sense because if a consumer feels more confident they will be more inclined to spend money on health care services, such as dental and physician services.Chart 5.2

Note: **means the variable was not significant. Computation done by the author.







Adjusted R-Square

Root MSE

1970Q1-2014Q4

2.04671(<.0001)

0.38256(<.0001)

** ** ** ** ** 0.1392 2.43993

1980Q1-2014Q4

1.70738(<.0001)

0.38065(<.0001)

** ** ** ** ** 0.1332 2.07160

1990Q1-2014Q4

1.58642(<.0001)

0.42704(<.0001)

** ** ** ** ** 0.1614 1.77050

2000Q1-2014Q4

-1.26892(0.4512)

0.26352(0.0562)

** ** 0.04305(0.0455)

** ** 0.1471 1.91498

Consumer confidence for the most part doesn’t have any predictive power over the health care industry or at least very weak predictive power. The lack of significance shows that no matter how consumers feel about their economic well being they are willing to do anything to purchase health care services. This makes sense because health care services for the most part are necessary. Dental services could be argued as more of a want than a need. A person can take care of their teeth by themselves and don’t necessarily need a dentist to maintain there teeth, unless of course you have a terrible tooth ache and the only way to fix it is by going to the dentist. Also, physician services also could be weakly argued as a want as well. Plastic surgery falls under this category and that by most means is a want and consumer confidence could play a role in the purchase of said services. Also, certain types of surgeries could be deemed not necessary and if say a consumer feels confident about their finances they could take the liberty in doing reconstructive surgery, on a knee for example. Unless of course the surgery was absolutely necessary, which is common in most cases. People in general aren’t going to want to have surgery unless they absolutely need it. These arguments could possibly explain why consumer sentiment was proven significant at the two quarter lag like previously discussed above. These wants could have been the driving force behind the significance of sentiment.

The other components of health care services can most definitely be defined as needs and in no way can consumer confidence shape the decisions for the purchases of these services. Paramedic services are usually in emergency situations and consumer’s sentiment wouldn’t affect the purchase of this service. Hospitals and nursing home services are also for the most part purchased in times of necessity. No person I know would go to a hospital and use their services unless they absolutely had to and therefore how a consumer feels about their economic well being would not play a role in the decision to purchase hospital services. Nursing home services could be argued as a service that is wanted by relatives of older persons because the older relatives can’t live on their own anymore. However, the move to a nursing

home may be some people’s only option and therefore sentiment won’t shape these decisions.

The next service variable up for discussion is recreational services. These types of services are comprised of memberships to clubs, parks, sport centers, and museums. Other components of the recreational services industry include gambling services, audio-video, photographic, and information processing equipment services, as well as veterinary services. Chart 6.1 graphically shows the relationship between consumer sentiment and recreational services.

Chart 6.1

Looking at the recreational series, it appears to have large oscillations throughout, with large peaks and troughs for the most part. The biggest jump that is observed is around 1991 where there is a large decline in the use of recreational services, and then a large spike only a few years after. It doesn’t appear that consumer sentiment follows similar paths as recreational services. Though, there appears to be declines in sentiment in 2000 and 2008 and at the same time there are declines in recreational services. This could mean that sentiment in the current quarter will affect recreational services in the current quarter. So, the use of lagged consumer sentiment may not be significant. To further analyze, chart 6.2 displays the statistical results.

Chart 6.2Note: **means the variable was not significant. Computation done by the author.







Adjusted R-Square

Root MSE

1970Q1-2014Q4

2.40970(<.0001)

0.43455(<.0001)

** ** ** ** ** 0.1843 3.76645

1980Q1-2014Q4

1.85536(<.0001)

0.51790(<.0001)

** ** ** ** ** 0.2629 3.52885

1990Q1-2014Q4

-0.87534(0.7086)

0.38006(<.0001)

** 0.16793(0.0004)

** -0.1378(0.0029)

** 0.2724 3.16462

2000Q1-2014Q4

1.00831(0.0227)

0.36833(0.0036)

** ** ** ** ** 0.1218 2.96670

The auto-regressive terms prove significant at a one quarter lag length only. So, for every time period an auto-regressive model is implemented. Now that all possible past information is being used in the model, consumer sentiment must be tested to see if it has any additional predictive power. For all time periods except 1990-2014 consumer sentiment proves to be insignificant. The time period of 1990-2014 has two lags of consumer sentiment that prove significant. It is very interesting though because the one quarter lag has a parameter value that is positive but the three quarter lag is negative. This could mean that when consumers are feeling confident in the short run they choose to purchase more recreational services, but when they feel confident in the long term they chose not to purchase the service. Observing the adjusted r-square value for 1980-2014 and 1990-2014 there appears to be only a slight difference. When the consumer sentiment variables were included in 1990-2014 and proven significant they only slightly increased the adjusted r-square value. This statement conveys that maybe the additional explanatory power of consumer sentiment is very minimal. Another interesting observation is that the adjusted r-square is the lowest for the time period of 2000-2014. As of recent times it appears that the auto-regressive term of recreational services explains less about current recreational service purchases.

The lack of significance of the lagged consumer sentiment variables proves that they don’t provide any extra explanatory power for recreational services. This is the opposite of what I would have thought for this particular service variable. Recreational services are wants for the most part and are not necessarily must haves for anyone. Therefore, it would be thought that the confidence of consumers would shape their decision making in the purchasing of these particular services. The only component of this service variable that could be regarded as a need is veterinary services. When a pet is sick, a consumer must bring it to the vet and has no choice. However, other components of the recreational services variable, like gambling, theaters, museums and sport facilities are facets of the entertainment

industry. Entertainment is strictly a want for people and consumer confidence would be thought to play a role in how consumers purchase these goods. Nevertheless, the statistics show that after accounting for all possible information, that is included in lagged dependent variable, lagged consumer sentiment doesn’t granger cause recreational services.

Another aspect of the recreational services is audio-video services. This means cable, satellite and radio services. These goods are in almost every household in America, unless the consumer is extremely poor. They have become second nature to almost every American and therefore the purchase of these goods will not be affected by how consumers feel about their economic well being because they will more than likely be purchased regardless. Though lagged consumer sentiment doesn’t appear to have any forecasting ability for recreational services, it could be the case that current consumer sentiment affects current purchases of the recreational services.

The next service variable that will looked at is food services. This type of service makes up food provided and purchased for school lunches. It also includes food for employees including military, and food accommodations for hotels and motels. Chart 7.1 will depict the visual relationship between consumer sentiment and the food service industry.

Chart 7.1

The food series appears to be very volatile, especially up until 1992. The oscillations are very large, but then appear to get smaller after that. Additionally, it doesn’t seem like there is any sort of relationship between these two series. However, around 2008 it appears that when consumer sentiment drops, the food series also takes a dip. Chart 7.2 will help clarify this.

Chart 7.2Note: **means the variable was not significant. Computation done by the author. * means that the variable was significant at the 10% level







Adjusted R-Square

Root MSE

1970Q1-2014Q4

-1.52458(0.5103)

** ** 0.04792*(0.0800)

** ** ** 0.0116 4.50100

1980Q1-2014Q4

-3.69776(0.0965)

** ** 0.06820(0.0085)

** ** ** 0.0423 3.82222

1990Q1-2014Q4

-1.26978(0.6413)

** ** 0.14175(0.0029)

** ** -0.10486(0.0258)

0.0713 3.68422

2000Q1-2014Q4

1.19408(0.0141)

0.37124(0.0027)

** ** ** ** ** 0.1304 3.17623

The auto-regressive terms for the first three time periods are not significant, but for the final time period an auto-regressive variable of one quarter lag length proves to be significant. When the auto-regressive terms are not included because of lack of significance it appears that consumer sentiment is significant in explaining food services. This is similar to the case previously discussed about the housing and utilities service variable. When an auto-regressive term is included like in the case of 2000-2014, consumer sentiment becomes insignificant. It is interesting that for the preceding three time periods an auto-regressive term is never significant, but as of recent times a one-quarter lag is significant. Also, when just an auto-regressive term is included like in the case of 2000-2014 the adjusted r-square is the largest, compared to the other three cases where sentiment proved significant. This illustrates that an auto-regressive term is more effective at explaining the dependent variable.

In the cases where consumer sentiment proved significant the parameter values for the most part are positive. Meaning when confidence increases the purchase of food services increases. The only objection to this is in the case where a four quarter lag proved significant. This parameter value is curiously negative.

Food services encompass any sort of food service for any educational, occupational or military facility. Food is the driving force behind human life and is a must have for every human being. Hence, one could conclude that food must be purchased no matter how you feel about your economic well being. Interestingly

enough, it proves that for at least the first three time periods tested that consumer confidence does have forecasting ability over the purchasing of food services. Maybe confidence plays a role in the quality of the food that is purchased. Lack of confidence could mean that less expensive food options are purchased. Nevertheless, as of recent times like the 2000-2014 test, autoregressive variables prove to be more useful than consumer sentiment at predicting future food service purchases.

A perplexing find is that in the cases where an auto-regressive term proves insignificant the consumer confidence variables prove significant and vice versa. As previously discussed this is the same exact thing that happened with the housing and utilities variable. Without an auto-regressive term in the model it is hard to truly assert the claim that consumer confidence can be used to predict food services because granger causality methodology is not being enforced. There could just simply be a correlation between these two variables and not truly causation. So, we must be careful when stating if consumer confidence can predict the food or housing and utilities service industry.

The final service variable that will be looked at is other services. This variable is formed by every other service not encompassed by the previously discussed service variables. This variable is comprised of communication services such as telecommunications, Internet access, postal and delivery. It also is made up of education services like college, nursery, elementary, and secondary schools. This variable is also made of professional services like legal, accounting, funeral/burial services, social services/religious activities and finally household maintenance. Chart 8.1 shows the graphical representation.

Chart 8.1

The other series doesn’t appear that volatile and it seems to follow similar paths as the consumer sentiment series. This leads to the conclusion that consumer sentiment could play a role in the majority of the micro components of this very large service variable. To further break down this relationship chart 8.2 is presented.

Chart 8.2**means the variable was not significant. Computation done by the author. * means that the variable was significant at the 10% level







Adjusted R-Square

Root MSE

1970Q1-2014Q4

-4.64338(0.0264)

0.19952(0.0103)

0.14117*(0.0663)

0.07534(0.0042)

** ** ** 0.1910 3.61394

1980Q1-2014Q4

-4.00381(0.0970)

0.17847(0.0403)

0.17574(0.0435)

0.06734(0.0234)

** ** ** 0.1761 3.67412

1990Q1-2014Q4

-7.78888(0.0040)

0.12712***(0.2222)

** 0.11406(0.0005)

** ** ** 0.1839 3.48452

2000Q1-2014Q4

-5.57657(0.0348)

0.22147*(0.0930)

** 0.08338(0.0126)

** ** ** 0.2445 2.58987

***means variable became insignificant after addition of lagged consumer sentiment

The one quarter auto-regressive terms proved significant for every time period, even though only weakly significant at the 10% level in the 1970-2014 and 2000-2014 periods. Also, the first two time periods have a second auto-regressive term that was significant as well, even though only weakly significant. Something occurred with the inclusion of lagged consumer sentiment for the 1990-2014 time period. Consumer sentiment proved significant when included but it turned the auto-regressive term to become insignificant. Nonetheless, lagged consumer sentiment was included anyway. A one quarter lag of consumer sentiment proved significant for every time period and there was a significant auto-regressive term in every model for the most part. Also, the adjusted r-square for the most recent time period is the largest. This could mean that consumer sentiment as of recent times can explain more of the variation in this other service variable and is becoming more prevalent. The parameter estimates all are positive, implying that increased consumer confidence increases the purchase of these services, which is to be expected.

Lets further break down this service variable to try and understand where the consumer sentiment significance is coming from. The communications component entails telecommunications and Internet access, both of these are things that people in the 21st century use on an everyday basis so therefore consumer confidence probably plays a minuscule role in the purchase of these services.

Furthermore, postal service is something provided by the government and has been around since the formation of this country so it would be hard to argue that consumer confidence will play a role here. The one aspect of the communications branch that consumer sentiment could affect is the delivery services. Delivery services are strictly dependent upon how much people consume and order products to be shipped to their house. Naturally, consumer confidence will play a role in how many durable goods consumers buy. So, if consumer confidence is low they will choose to purchase less goods and the need for delivery service will decline. There could be a possible link there.

The next component of this service variable is educational services. This is made up of nursery, elementary, and secondary schools. The latter two are requirements by the government and it is against the law to not attend these schools. So, this part of educational services can’t be affected by consumer confidence. The education service also encompasses college. Now, law does not require college attendance and where a student decides to go is entirely up to them. Of course all colleges are not the same and some are more expensive than others. You can choose private or public colleges depending on a variety of factors. One of the main driving forces in the student’s decision making is how much money the tuition and living expenses would be. Therefore, if a consumer is feeling confident about there current and future economic conditions they will more then likely decide to go to a more expensive school, simply because they can afford it. Paying off college loans is becoming more and more of a dilemma for students everywhere as of late, and therefore if a person didn’t think they could pay them off they wouldn’t attend that certain school. Hence, it appears, at least for the college component of the education services, that consumer confidence will affect decision making.

The final feature of this variable is professional services. Legal services could necessarily be something that a consumer wants if they are to get in trouble with the law. People of course have the right to represent themselves even though this is not recommended. The use of lawyer’s services usually helps the person in time of need so regardless of the price a person will usually opt to purchase a lawyer. Nevertheless, some lawyers are more expensive then others so if someone doesn’t feel confident about their finances they could decide to hire a cheaper lawyer. Consumer confidence in this instance plays a minor role in their decision. Accounting services are a luxury as well, if you don’t enjoy doing your taxes or need someone to manage your debits and credits you could hire an accountant. Despite this, accounting services may be absolutely necessary if you don’t understand how to report your taxes to the government. Therefore these services could possibly be a necessity in some people’s cases but consumer confidence could definitely play a role in the decision to hire an accountant.

Burial and funeral services are something that are absolutely necessary if someone passes away, there is no way around this. Unless of course you want to go dig a hole in the backyard to bury your loved ones, but this is obviously not an option. So in this particular case consumer confidence doesn’t affect this service. Religious activities and social service are usually things that people contribute towards because they have faith in an upper being and also because they like helping people. Regardless how these people feel about their finances, they will

more then likely always decide to contribute to these services because it is an inherent duty in their mind.

The final aspect of this variable is household maintenance. The BEA defines this as domestic cleaning services. If someone decides that they are economically comfortable enough they will decide to purchase this luxury service. Having someone clean your house is by no means a necessity, of course you can always just clean your house yourself if need be. This service is for more affluent people who don’t have the time to keep their house clean the way they want it. So, if a consumer is feeling like they can bear the expense to have someone simply clean their house, they will take on that luxury. Therefore, household maintenance is something that will be affected by consumer confidence.

For every time period an auto-regressive model was used and lagged consumer confidence was significant, thus it can be concluded that consumer sentiment granger causes this other service variable. It can be concluded that lagged consumer sentiment of up to one quarter can be used to forecast this service variable.

Conclusion

The growing service sector in the United States has gotten more attention as of late because of how much it contributes to the United States GDP. The service industry contributes to almost 50% of the total GDP in this country. Now more then ever it would intrigue anyone to try and understand the driving factors behind the service industry. If one was able to predict the future value of the services sector it would provide great information on how well the economy is going to do as a whole. Past papers have discussed and proven that consumer sentiment does in fact affect the service industry. No one though has dug deeper into the service industry to further understand what sectors of the service industry are actually affected by consumer sentiment and which sectors are not. So, in this paper the main seven sectors of the service industry are tested. This is so we can see which components of the service industry are actually affected by sentiment.

As previously proven lagged consumer sentiment was effective in forecasting the service industry as a whole. An auto-regressive model was implemented and through granger causality it is proven that consumer sentiment does have a casual relationship with the service industry. An auto-regressive model could not be used for only two of the seven sectors. The housing and utilities sector showed that auto-regressive terms were insignificant. This is a strange result, because past values of the same economic variable usually do quite well at predicting future values of that same variable. That is why I found it odd that for this variable an auto-regressive term was insignificant. Additionally, one quarter lagged consumer sentiment was significant at predicting the service variable. The absence of auto-regressive terms in the model make it difficult to assert the claim that consumer sentiment has any sort of forecasting power or causal relationship on the service variable in question. Another interesting point is in the case of the food service sector. When an auto-regressive term was in the model and proven significant, consumer sentiment was

proven to be insignificant. Opposite of that result, when an auto-regressive model wasn’t used, consumer sentiment was then significant. The implementation of an auto-regressive term is absolutely necessary though if a strong causal relationship were to be proven. The one case where an auto-regressive model was used for every time period and consumer sentiment still proved to be significant for every time period was the “other” service variable. There can be strong conviction that there is a causal relationship between consumer sentiment and this particular service sector.

All other service sectors proved that when an auto-regressive term was in the model, consumer sentiment provided little to no additional information on the future value. The auto-regressive terms were for the most part strongly significant and provided much information on future values of that said variable. This could be because a lot of these services are necessities that people need to either survive or get through their everyday life. When a service is a need and not a want, consumer confidence can’t play a role in the decision making process because the service must be purchased with no exception. In terms of services that people don’t essentially need, consumer confidence could shape a consumers decision.

Further considerations for testing these relationships would be to continue to break down the seven main sectors into their individual components and see which of those individual components are affected by consumer sentiment. There are many different sections of the service industry and therefore more testing can be done to further test these claims. Also, instead of using lagged consumer sentiment, one could regress current service variables on current sentiment. Maybe the sentiment a consumer feels in the current quarter will affect their decisions today rather then tomorrow. So, in terms of trying to forecast the service industry it may be interesting to see if current sentiment can be forecasted and then used to forecast the current service output. A final consideration for further analysis would be to use the expectations component of the consumer sentiment index, instead of the overall index value. Nevertheless, the service industry is growing by the day and the use of many different economic variables other then consumer sentiment should be tested for forecasting ability.

References

Amadeo, Kimberly. "What Are the Components of GDP?" About News, 2015. Web. 12 May 2015. <http://useconomy.about.com/od/glossary/g/Gdp-Per-Capita.htm>.

"Charts-The Index of Consumer Sentiment." Survey of Consumers. University of Michigan, 1 June 2015. Web. 19 Apr. 2015.<http://data.sca.isr.umich.edu/tables.php>.

Diebold, Francis X. "Forecasting with Regression Models." Elements of Forecasting. Third ed. Michael B. Mercier, 2004. 247-264. Print.

"Growth of the Service Sector." World Bank, 2009. Web. 6 June 2015. <http://www.worldbank.org/depweb/beyond/beyondco/beg_09.pdf>.

Ludvigson, Sydney C. (2004), “Consumer Confidence and Consumer Spending,” The Journal of Economic Perspectives, Vol. 18, No. 2. pp. 29-50

"Table 2.3.1. Percent Change From Preceding Period in Real Personal Consumption Expenditures by Major Type of Product." U.S Department of Commerce-Bureau of Economic Analysis. Bureau of Economic Analysis, 24 June 2015. Web. 20 Apr. 2015. <http://www.bea.gov/iTable/iTable.cfm?ReqID=9&step=1#reqid=9&step=3&isuri=1&903=61>.

"The Components of Consumption." Boston College, 2012. Web. 9 Apr. 2015. <https://www2.bc.edu/~murphyro/EC204/Supps/SuppCh16.pdf>.

Wooldridge, Jeffrey. "Advanced Time Series Topics." Introductory Econometrics: A Modern Approach. Fourth ed. Cengage Learning, 2009. 649-652. Print.

Masters Thesis

Documents