Real estate rent seeking : examining the ability of real estate agents to use occupational licensing laws to earn economic rents

(1)

R

EAL

E

STATE

R

ENT

S

EEKING

?

Examining the ability of real estate agents to use

occupational licensing laws to earn economic rents

Faculty of Economics and Business Amsterdam School of Economics

University of Amsterdam Austin Harris

10824871

Supervisor: Jo Seldeslachts

This thesis is submitted for the degree of Master of Science in Economics with a Specialisation in Industrial Organisation, Regulation and Competition Policy

August 2015 Amsterdam

(2)

ii ri ed m an & M o rg an 2 0 14

A

BSTRACT

This thesis examines the effect of occupational licencing on the U.S real estate agent market, in particular testing whether agents who work in states with higher educational requirements are less numerous or earn more than those in states with lower requirements. It adds to the small amount of existing literature on this topic by using newly collected data as well as applying difference techniques such as random effects and difference in differences models. It finds evidence that higher educational requirements to become a real estate agent are causally linked to higher agent earnings and fewer agents per capita after controlling for appropriate variables. These results provide further evidence of the potentially negative effect of occupational licensure on consumer welfare.

(3)

C

ONTENTS

1 INTRODUCTION ... 4

1.1RESEARCH QUESTION ... 6

1.2THESIS STRUCTURE ... 6

1.3BACKGROUND ... 7

1.3.1 Occupational Licensure in Practice ... 7

1.3.2 Operation of the Real Estate Industry in the U.S ... 9

2 LITERATURE REVIEW ... 10

2.1THEORETICAL... 10

2.1.1 For ... 10

2.1.2 Against ... 12

2.2EMPIRICAL STUDIES ... 13

2.3REAL ESTATE SPECIFIC LITERATURE ... 14

2.4GAPS THIS THESIS WILL FILL ... 16

3 METHODOLOGY AND DATA ... 18

3.1OVERVIEW OF THE METHODOLOGY ... 18

3.1.1 Illinois and the 2011 changes to the Real Estate Licensing Act... 21

3.2THE MODELS ... 22

3.3THE DATA ... 23

3.3.1 Summary Statistics ... 25

3.3.2 Data Shortfalls ... 27

4 RESULTS ... 28

4.1RANDOM EFFECTS MODEL RESULTS ... 28

4.1.1 Agent Earnings ... 28

4.1.2 Agent Numbers per Capita ... 32

4.2DIFFERENCE IN DIFFERENCE CASE STUDY RESULTS ... 36

4.2.1 Agent Earnings ... 36

4.2.2 Agent Numbers per Capita ... 39

5 DISCUSSION ... 43

5.1MAIN RESULTS DISCUSSION ... 43

5.2CASE STUDY RESULTS DISCUSSION ... 44

5.3POLICY IMPLICATIONS ... 45

5.3.1 Policy Recommendations ... 46

(4)

6.1AREAS FOR FUTURE STUDY ... 48

(5)

L

IST OF

T

ABLES

TABLE1.1OUTLINE OF COMMON LICENSURE FORMS ... 7

TABLE 3.1VARIABLE DEFINITION ... 23

TABLE 3.2DEPENDANT VARIABLE SUMMARY ... 25

TABLE 3.3EXPLANATORY VARIABLE SUMMARY ... 26

TABLE 3.4CONTROL VARIABLE SUMMARY... 26

TABLE 4.1AGENT EARNINGS REGRESSION RESULTS... 29

TABLE 4.2NUMBERS OF AGENTS REGRESSION RESULTS ... 33

TABLE 4.3EARNINGS DIFFERENCE IN DIFFERENCE REGRESSION RESULTS ... 37

(6)

1 I

NTRODUCTION

The artificial restriction of entry into particular occupations has a long history stretching from medieval guilds to today’s more modern occupational licencing regimes. Proponents of occupational licencing argue that by requiring new entrants to demonstrate a minimum level of skill or proficiency, these restrictions ensure minimum acceptable quality standards within the industry and serve to protect consumers from shoddy fly by night sellers. They argue that because so many fields require specialist knowledge to properly assess, the average member of the public is not well placed to compare providers. Opponents of these restrictions argue that entry restrictions are used to artificially restrict competition and drive up the earnings of protected insiders. Instead, they argue that while some standards are required, licenced professions have a vested interest in increasing these standards beyond what is reasonably required. The dual effect on both quality and quantity of such licencing means that their use is open to potential abuse by industry bodies seeking to earn economic rents.1

Licencing is playing an increasingly important role in the economy, therefore understanding its effects are necessary. In 1950 about 4.5% of the workforce required a licence to practice. By 2000 the percentage was 20%; by 2013 the percentage had grown again to 35%. Being licenced appears to be highly beneficial, with studies showing that

1_{A note on terminology: When referring to both salespersons and brokers together, this thesis} will use the term agents or real estate agents. Otherwise the various categories will be referred to by name.

(7)

licencing in the U.S is associated with a wage premium of around 18% (Kleiner, Krueger 2013).

Despite licencing’s growing prominence in the economy, this topic is not widely covered in the academic literature. With the notable exception of Morris Kleiner there are few recent articles on this topic in the top economic journals. This thesis aims to make a contribution to the field by providing some discussion of licencing using very recent data.

This thesis will study the impact of occupational licencing by analysing its effect on the earnings and numbers of real estate agents in the United States. In the U.S, real estate agents are licenced at the state level. Between states there is a wide variety of standards and requirements which potential agents must meet before they can be licenced, which at face value does not appear to be warranted by the different market conditions within each state. For example in Texas an individual wishing to earn their real estate agent’s licence is required to undergo 210 hours of coursework before sitting an exam; By contrast, Alaska requires only 24 hours of instruction. Similarly to become a broker (Agents typically must be supervised by a broker) Texas requires an additional 630 hours of coursework over and above the salesperson requirements. This compares unfavourably to the 8 hours required by Nebraska.

While each state has slightly different laws and customs, it is unlikely that these differences require such variation between states. Anecdotal evidence suggests that there is a link between higher educational standards and a reduction in the number of agents, for example Jeff Foster the Deputy Director of the Colorado Real Estate Commission was quoted saying:

“We sure saw a drop off in our applicants because we did raise the bar and there were less applicants” (Evans, 2002)

Previous studies of other licenced industries, particularly dentistry (Kleiner & Kudrle, 2000) and opticians (Tmmons & Mills, 2015) have shown that states with higher educational requirements do not provide higher qualities of service as measured by

(8)

complaint volumes. This divergence therefore appears to be the result of concerted efforts by the industry to restrict entry into these state level markets. These efforts impose serious costs with one study of the cost of higher education standards for brokers estimating the cost to American consumers at 5.4 billion a year (Barker 2008). Testing this assertion and measuring its potential effects provides the motivation behind this thesis.

1.1 Research Question

This motivation clearly leads into the Research Question for this thesis, which is laid out below.

Are occupational licencing requirements in the U.S being used as a tool to raise real estate agent earnings or to reduce the number of agents per capita? If so how effective is it?

This question covers a number of issues. It addresses salespersons and brokers separately, as in many states the barriers to entry for brokers are far higher than that for salespersons.

1.2 Thesis Structure

The remainder of this thesis is set out as follows. The rest of Chapter 1 will provide the reader with background information about the real estate industry in the U.S and how it operates necessary to place the analysis in its proper context, as well as a summary of occupational licensing in practice. Chapter 2 will outline the existing academic literature on licencing and show where this paper can make a positive contribution to the ongoing debate on the effect of licensure. Chapter 3 will discuss the methodology and data to be used in assessing the research question. Chapter 4 contains the results of the regression analysis while Chapter 5 will provide analysis on its meaning as well as discuss possible policy implications of these findings. Lastly Chapter 6 concludes and provides readers to possible extensions to this analysis.

(9)

1.3 Background

This section presents an overview of key background information on areas central to this paper. This includes a brief outline on different types of occupational licencing, a review of the real estate industry and a primer on how the real estate market in the U.S functions.

1.3.1 Occupational Licensure in Practice

Broadly speaking, there are three types of occupational licensure, Registration, Certification and Licensure. The differences between these three categories are outlined in Table1.1below:

Table1.1 Outline of common licensure forms

Type Description Example

Registration

Individual registers with the government which places their name on a public list of practicing individuals

Dietitian, Architect

Certification

A Non-Governmental Organisation has certified that the individual in question has achieved a certain level of competency. Individuals can advertise this fact; however non certified individuals can still perform these tasks as long as they don’t claim certified status.

Certified Financial Analyst, Board Certified Behaviour Analyst, Travel Agents, Car Mechanics

Licensure

Government requires a licence issued either through an NGO or the government itself in order to practice. Individuals are required to demonstrate their capability before being issued a licence. Receiving compensation without the relevant licence is illegal.

Doctors, Surveyors Lawyers, Real Estate Agents

(10)

Governments can decide to introduce licencing of an occupation for a number of reasons. The U.S Council of State Government outlined in their 1952 volume “Occupational Licencing Legislation in the States, the general rationale for and against regulating occupations.

The key benefits are as follows:

1. Given increasing urbanisation, only Governments can adequately determine whether an individual has the level of knowledge or competency needed to serve in a profession

2. Increasing specialisation of occupations means that the average individual can no longer tell the honourable and dishonourable practitioners apart – necessitating the state to step in

3. Licencing boards “provide the average citizen who has limited time and money with a swift, simple and inexpensive administrative avenue for redress” against malpractice, while board sanctions serve to hold licence holders to high standards

4. Licencing boards who are experts in the field are readily able to assess new technology and developments to determine what is suitable for the profession to use.

Against these benefits the volume outlines the following shortfalls:

1. Licencing bodies may limit the number of entrants into the profession, creating artificial scarcity.

2. These restrictions can restrict competition in the market, raising prices and wages

3. Licencing bodies have an incentive to increase the limits of a particular profession in order to capture additional work by excluding potential competitors.

4. Private associations which have the ability to revoke individual’s licences amount to a privatisation of regulatory powers to a self-interested body. (CouncilofStateGovernments1952)

(11)

1.3.2 Operation of the Real Estate Industry in the U.S

Although there are differences between states, the majority of agents fall into one of two categories, salespersons/agents and brokers.2_{Generally speaking salespeople}

handle the day to day interaction with clients, performing tasks such as soliciting listings, showing homes to prospective buyers and posting advertisements to sell houses. Brokers on the other hand supervise salespersons actions, signoff on sales and provide the branding, advertising and back office functions to assist in closing sales. Most states require that salespersons be affiliated with or work directly for a broker.

Real estate agents in the U.S work for both buyers and sellers of property. Sellers who use a broker3_{to list their property enter into a listing agreement for a fee, which is}

typically a percentage commission of the sale price. These brokers will list the property on the Multiple Listing Service (MLS) which is an online database accessible only by agents which lists details about houses for sale in a given area. Buyers Brokers also known as cooperating Brokers, serve a similar function but for buyers. They will source homes for inspection, assist buyers in organising finance and advise buyers on making offers. Cooperating Brokers are paid by the seller’s broker who offers an unconditional offer of compensation on a home’s MLS listing to any broker that is the procuring cause of sale. Generally the two brokers will share the percentage commission, with the exact split being determined by market conditions. (Federal Trade Commission , 2007)

The salesperson responsible for the sale will pay their broker a proportion of the commission out of which the broker pays office overheads and generates profit. If there is no buyers broker the selling broker takes the full commission. If the property is being sold by the owner without a broker, known as ‘For Sale by Owner’ or FSBO, then the owner will have to negotiate with any buyers broker their commission. According to the NAR only 9% of houses in the U.S are sold in this manner (National Association of Realtors , 2014).

2 _{Different states may use different terminology but the forms at functionally similar.} Alternative names include associate broker, managing broker and principle broker. Exceptions include Colorado which has only a single level of licensure - brokers

(12)

2 L

ITERATURE

R

EVIEW

As mentioned in the introduction, formal restrictions on entering certain occupations have been a feature of the world for centuries. During this time various competing schools of thought have arisen, with the key point of difference being whether these restrictions are mostly helpful or harmful. This chapter will outline the key debates about occupational licencing in the literature, noting where there is general consensus and where there are open questions. It will do this by examining historical and theoretical analyses of licencing followed by key empirical studies and finish with a discussion of papers covering real estate licencing. It will conclude by discussing how this paper fits into the established body of work and where it aims to make a valuable contribution.

2.1 Theoretical

There is an ongoing debate within the economic literature on whether occupational licencing has positive or negative welfare effects. Arguments for and against licencing typically depend on the model chosen and which effect dominates. The following sections provide a short discussion of these main arguments.

2.1.1 For

Proponents of occupational licencing have generally relied upon models of adverse selection or moral hazard. Akerlof’s 1970 paper Market for lemons lays the foundation in this field. Akerlof argues that there is asymmetric information between buyers and

(13)

sellers in the types of markets which are typically licenced. In Akerlof’s model there are low and high quality goods, with high quality goods being of greater value to the buyer but costing the seller more to produce. Only the seller knows the quality of the good on offer and has every incentive to advertise a low quality good as high quality. Akerlof shows that in such a model the equilibrium price is such that high quality goods are pushed out of the market and only low quality ‘lemons’ remain in the market. In such situations Akerlof argues that this adverse selection results in market failure and that quality regulation can improve welfare outcomes (Akerlof, 1970). Leland (1979) expands Akerlof’s model and applies it to entry into specific occupations to show that licencing regulations can improve welfare outcomes. Similar to Akerlof, Leland’s model features high and low quality services and consumers who cannot tell the differences between the two before purchase. Without a differentiation of price by quality, higher cost high quality providers are driven out of the market, leaving only low quality providers. Leland argues that this adverse selection program can be overcome by restricting the quality of the occupation through licencing controls. These minimum standards disproportionally raise the costs of the lowest quality workers (who must put in more effort for a given quality compared to high quality workers) causing them to exit the market. This exit improves the equilibrium quality of workers in the occupation, lowering the adverse selection problem and resulting in an improvement in welfare outcomes (Leland, 1979).

An alternative to the adverse selection model proposed by Leland is the Moral Hazard model first outlined by (Shapiro, 1986). Shaprio makes a number of important changes to Leland’s model. Firstly he does not treat quality as given but rather direct function of the level of human capital investment or training. This investment has the effect of lowering marginal cost of providing quality. Licencing then takes the form of minimum levels of training rather than required standards of output. Secondly Shaprio relaxes the assumption that seller’s quality isn’t visible, allowing buyers to observe a seller’s quality with a lag.

In Shapiro’s model raising licencing requirements promotes the provision of quality service by lowering its cost and raises the market share of high quality workers. This reduces the incentive for the worker to cheat and provide low quality services. Assuming that consumer’s value this quality sufficiently, Shapiro argues that licencing will therefore lead to an improvement in overall welfare, even as some buyers are priced out of the market.

(14)

2.1.2 Against

The main argument in the literature against occupational licencing is that it raises prices and reduces competitive pressure in the market. Adam Smith first pointed out this tendency in Wealth of Nations where we commented that licencing was a way to “limit the number of apprentices per masters, thus ensuring higher earnings for persons in these occupations (Smith, 1994 [1776]) Friedman argues that because, for example, the best placed person to judge the quality of a plumber is another plumber, that occupational licencing leads to professions regulating themselves. He questions whether these licencing boards are truly “unbiased gatekeepers” or whether instead do they use the power to regulate entry to limit competitive pressures. Friedman also notes the tendency for regulated professions to expand the range of their tasks or entry requirements so as to further limit competition (Friedman, 1962). This argument for self-expanding professions is supported by (Gellhorn, 1976) who observes the growth of irrelevant requirements as barriers to potential new entrants.

The lack of unbiased gatekeepers matters as the benefits of regulation are concentrated in a relatively small group of individuals while the costs are diffused across the economy as a whole (Kleiner, 2006). Kleiner argues that this unequal distribution of costs and benefits incentivises active lobbying by industry to further entrench favourable regulation.

(Stigler, 1971) applies this line of argument to demonstrate that industries use regulation to seek one (or more) of four things:

 Direct Subsidies

 Control of entry of rivals

 Influence over the marker for substitutes and complements

 Price fixing

These four issues form the way in which industries benefit from regulation. Stigler provides a number of historical examples where industries used political influence to achieve control over one of the above

Stigler then develops a model which demonstrates the key characteristics of an occupation which should influence its ability to secure political power in the form of favourable regulations. These characteristics are;

(15)

 Size of the occupation – with larger occupations having more sway

 Per capita income of the occupation – larger incomes influence the potential rewards for political action

 Concentration of occupation in large cities – higher levels of concentration reduces the likelihood of free rider problems occurring and make it easier for regulation to occur at lower government levels i.e. city or county level

 Cohesive opposition to licencing – smaller customer groups are more likely to organise against licencing as the costs are born by a smaller group. All things being equal a smaller group is easier to mobilise

Both the for and the against arguments rely on certain assumptions about how the world works. Therefore in order to test these models to determine the practical effect of occupational licencing, real world data is required. Section 2.2 below discusses some of the empirical work aimed at testing the overall effect of licencing on consumer outcomes.

2.2 Empirical Studies

The empirical body of work on licensing typically follows one of two methodologies. The first involves conducting a macro level study aimed at understanding the prevalence of licencing in the economy as a whole, and possibly calculate its effect. Studies such as (Kleiner & Kudrle, 2000; Kleiner & Krueger, Analyzing the Extent and Influence of Occupational Licensing on the Labor Market, 2013) have found that licencing typically carries a significant cost to consumers with mixed results on the improvement in the quality or availability of services.

An alternatively methodology is to study specific industries as case studies in order to allow examination of specific issues in more depth. These reviews, such as (Tmmons & Mills, 2015; Shepard, 1978) assess whether consumers or the industry in question benefits from the regulation. This section will discuss work that fits into these two categories while the next looks at real estate specific research.

The most comprehensive study into the prevalence of occupational licencing in the U.S is a 2013 paper by Kleiner and Kruger. They use a national labor force survey

(16)

conducted on behalf of the authors in 2008 to show that 35% of American workers require either a license or certification to perform their jobs. The authors find that licensing is associated with 18% higher wages (Kleiner & Krueger, 2013). An earlier work by one of the authors found that the gains from licencing are greater for higher educated workers (Kleiner, 2000).

This increase in earnings from licensed professions has also been found in papers which have focused on specific industries. (Timmons & Thornton, 2010) examined barbers in the U.S using a standard human capital earnings function and found that states which required an apprenticeship period (roughly doubling the learning requirements) saw earning increases of 22% and a number of barbers per capita decrease of 25% compared with non-apprenticeship states. A similar study by the same authors into the massage industry found earnings increases of 16% in heavily licenced states (Thornton & Timmons, 2013).

Other studies have examined whether the increased training or stricter standards imposed by licencing bodies have led to improvements in quality outcomes for consumers. If so the negative consumer welfare effects of increases in prices could well be offset by quality increases. (Kleiner & Kudrle, 2000) Investigated the dental health of incoming U.S Air Force personnel to analyse the effects of varying levels of licensing strictness among different states. They found little evidence of a public benefit from stricter licencing as the increased cost of dental services was not matched with commensurate improvements in the quality of dental services provided. A similar study into opticians in the U.S found similar results; increases in price were not matched by any observable improvements in quality (Tmmons & Mills, 2015).

2.3 Real Estate specific literature

While some regulated industries, in particular medical fields, have been extensively studied, the real estate industry does not have a large body of work dedicated to it. There are only three papers published in academic journals in the last 30 years which explicitly address the issue of real estate agent licencing in the U.S, with two of those

(17)

using economic data from the 1980’s. This section will provide a short overview of these papers noting their methodology, findings and failings.

Simultaneous equations are used by each of the three existing papers on real estate licencing to build a model of the supply and demand for real estate agents. Each of the papers differ however in the data they had available, what they see driving demand and supply and lastly in the conclusions they draw.

(Johnson & Loucks, 1986) use a two stage least squares approach to assess the effect of entry barriers in the real estate industry on the number of licensees per capita, earnings and quality as measured by complaints. They find that there is no evidence that higher earnings arise from fewer brokers and that a reduction in the number of brokers due to lower exam pass rates results in an increase in the quality of service. They conclude that consumers benefit from industry regulation rather than agents, and that there is no evidence that increases in requirements is the result of the industry acting in its own interests. The authors acknowledge that their methodology suffers from a few notable weaknesses particularly the use of cross sectional data and the inability to differentiate between active and inactive brokers.

(Jud & Winkler, 2000) also examine the effect of examination pass rates and educational requirements on the numbers and incomes of agents using a two stage least squares methodology. They differ from Johnson and Loucks by using updated data, corrected for regional cost of living as well as directly linking earnings and the probability of passing the exam. In their model an individual decides to become an agent if Net Present Value of the investment in education and training, given the probability of passing the exam, is positive. They find that changes in the exam pass rate exert a significant influence on the supply of agents, but that educational requirements outside of the exam have a relatively small effect. Overall they determine that a 10% reduction in exam pass rates results in a 2.7% decrease in agent numbers and a 2% increase in agent earnings.

(Barker, 2008) approaches the issue of entry restrictions from an ethics stand point, questioning whether it is ethical for professionals to lobby politically for increased standards which only benefit themselves. Nevertheless the author determines that

(18)

entry restrictions have no positive effect on quality, but cost consumers $5.4 billion a year in increased fees paid to brokers. Like the previous two papers, Barker develops a two equation system for estimating supply and demand. Unlike the past two papers Barker uses panel data rather than cross sectional. Ultimately Barker finds that all three measures of licencing restrictions (pre licensure hours, continuing professional development requirements (CPD) and years of experience before becoming a broker) negatively affect supply, with years of experience and CPD being significant. According to Barkers results, an additional hour of annual CPD requirements raises broker incomes by $235 while an additional year of experience as a salesperson before becoming a broker increases income by $852.

2.4 Gaps this Thesis will fill

As noted above there is a dearth of studies covering the occupational licences, and specifically the real estate industry in the U.S. Two of the three papers published in this area use data from the 1980’s in their analysis. Therefore there is a clear space for this paper to revaluate the issue with updated data. Additionally this paper attempts to address short comings with each of the existing papers in this area. It does this by first using new, more detailed panel data to track variables over time rather than the more limited cross sectional data available to (Johnson & Loucks, 1986) Secondly it will test the effects of licences in a more granular manner by separately testing salespersons and the more exclusive broker licences unlike (Barker 2008). This is in order to test whether brokers have a greater ability to use licencing to influence earnings or competition compared to the lower ranked and more numerous salespersons. Lastly it considers all 50 states rather than a selection of Metropolitan areas as was used by (Jud & Winkler 2000), which will extend the validity of any findings.

This thesis also differs from existing methodologies in how it models the market. The papers discussed in the section above all attempt to model the supply and demand of agents using simultaneous equations to estimate supply and demand. However these models fail to accurately account for the specifics of the real estate market. The demand for brokers is unlikely to be substantially effected by the price of brokerage services, being far more likely to be driven by exogenous factors such as changes in employment

(19)

or family growth. The price of using an agent varies, but is generally a small amount of the total sales price, between 5.5%-6% of the final sales price spilt between the buyers and sellers agent. However, if only one half of the transaction users an agent, the commission is only halved. Therefore both parties would have to forego using an agent in order for price to fully influence demand for agent services.

Furthermore, price competition amongst agents has remained low according to a Federal Trade Commission study into brokerage services. While average commission percentages have trended downwards, this has been more than offset by increased house prices over the same period with the end result being that agents earn more per house than before (Federal Trade Commission , 2007). Surveys from the NAR indicate some 92% of real estate transactions in the U.S use brokerage services (National Association of Realtors , 2014). Therefore it is reasonable to assume that the demand curve for brokerage services is largely exogenous and focus on the supply side of the market.

(20)

3 M

ETHODOLOGY AND

D

ATA

This chapter will outline the research methodology and data used in this paper’s analysis of the real estate market. It will start by providing a short overview of the methodology and how it differs from the existing body of work discussed in the previous chapter. This is followed by a presentation of the models to be used in the regressions, expected effects of each variable and finally a discussion on the rationale behind various decisions. The second half of this chapter will cover the data being used in each of the regressions. It will start with summary statistics before going into greater depth about the sources and calculation of some of the key statistics being used.

3.1 Overview of the Methodology

Past papers on real estate licencing have developed simultaneous equations attempting to model the demand and supply of agents. However as was discussed in section 2.4 above the particular nature of the real estate market means that demand for agents can be considered largely exogenous while the supply of agents is more of an open question. Therefore this paper will instead look at the outcomes that regulation has on the potential supply of agents. In particular we will focus on two key measures, earnings and numbers.4_{We choose these because regulatory theory suggests that}

4_{For the purposes of meaningful comparison between states, agent numbers are all calculated} per 1000 population

(21)

where excessive entry restrictions are in place the normal causal loop between numbers and earnings is reversed. That is instead of higher earnings causing an influx of new agents, the artificial restriction in numbers results in higher earnings. The rationale for these divisions is simple. Earnings and level of competition5_{have been}

shown to be the most influenced by artificial restrictions imposed by occupational licences. Therefore if the real estate industry is using licensure to benefit industry insiders, these are the two areas where it would show the most.

This paper will also separate the analysis of salespersons and brokers. Brokers are more experienced agents that have undergone additional training and testing, and typically supervise salespersons. It is reasonable then to assume that brokers have more to gain than salespersons from artificial restrictions to competition. By separately testing salespersons and brokers this paper will be able to differentiate how licencing restrictions affect both types of agents and whether brokers are better able to benefit from increased regulation.

Therefore there are four key dependant Y variables which we will test in our regressions: salesperson earnings, broker earnings, salesperson numbers and finally broker numbers.

We will test these variables in two phases using two different types of regressions. The first phase will use panel data collected by the author covering the 50 U.S states from 2007-2014 to test each of the dependant Y variables with a random effects model which will test whether higher earnings or a lower number of licenses per captia is associated with higher educational requirements, after controlling for variables such as urbanisation and state economic activity or housing sales. The dates have been chosen partly because they cover the economic downturn which affected the U.S in the aftermath of the Global Financial Crisis, but also because the most recent paper which covered real estate licencing used data until 2006. This phase will form the general case from which the bulk of conclusions and policy recommendations will be drawn from.

(22)

The second phase will consist of a case study testing the four dependant Y variables in a difference-in-difference regression. It will use the state of Illinois as the treated variable and both the Midwest and U.S itself as the controls. Illinois roughly doubled the entry requirements to become a salesperson or broker halfway through the reference period. During the period in question, Illinois was the only state to change its entry requirements. Given this, Illinois’ changes offer a useful natural experiment for us to check whether the findings in the first phase are accurate.

The rationale behind using two different control groups in the second phase of regressions is to provide additional assurance that the findings are accurate. In effect using two separate control groups creates two separate tests from which to draw inferences from and from which to confirm the results of the main regression.

The Midwest is used as a control because it is the geographical and statistical unit which Illinois is placed in by the BLS. It is therefore, the closest available comparator to Illinois. The U.S as a whole is used as a separate, larger control group to provide a larger data set and therefore lower standard errors. Illinois as a state closely tracks the national average for the control variables being used in this regression. It contains a mix of urban, suburban and rural population centres which are representative of the makeup of the country as a whole. The saying “will it play in Peoria” refers to Peoria Illinois and is widely used in advertising to test whether a product or campaign will appeal to a middle America, mainstream taste.

In both cases we will control for biases in standard errors through the use of robust or clustered standard errors as appropriate.

In an ideal world a fixed effects rather than random effects model would be used to test these hypotheses. Unfortunately changes in licencing laws occur slowly, infrequently and over a much longer time period than data could be collected for this paper, effectively being time invariant. This necessitates the use of a random effects model instead. The shortcomings of the random effects model – the assumption that variation across entities is assumed to be random and uncorrelated with the predictor

(23)

or independent variables included in the model is controlled for in two ways. Firstly the possibility of omitted variable bias is minimised through the inclusion of a wide array of control variables covering economic growth, urbanisation, employment and other factors. Secondly by taking advantage of the natural experiment environment which occurred after Illinois changed their regulations; the proposed difference-in-difference regressions, which although based on a small treatment group, will still provide additional evidence and confirmation that the findings of the original regressions are accurate.

One issue that this paper will not examine is the effect of increased educational requirements on the quality of real estate brokerage services. This is for two reasons. The first is a lack of space and time to expand the current model to consider issues of quality. The second is a significant body of work which examines quality issues in licencing and finds that in many industries there is no improvement in quality (Kleiner & Kudrle, 2000; Peltzman, 1987; Shilling & Sirmans, 1988). Therefore, for the purposes of this thesis service quality will be treated as constant and uncorrelated with educational requirements.

3.1.1 Illinois and the 2011 changes to the Real Estate Licensing Act

On May 1st_{2011 a number of changes to the Illinois Real Estate Licensing Act of 2000}

came into effect. The salesperson licence was removed and a new category – Managing Broker – was established. Salespersons were required to undertake an additional 45 hours of education (to 90 total) or risk losing their license to practice. Brokers could either take no action and in effect be demoted or undertake an additional 45 hours (to 165 total plus a doubling of the CPD requirements) in order to become Managing Brokers and continue performing the same functions as before.

These changes offer a rare opportunity to observe the effects of occupational licencing laws in real time. By doubling the requirements to enter the real estate profession, Illinois has created the conditions for a natural experiment. It allows us to compare how the earnings and numbers of agents change before and after the amendments

(24)

came into effect against other states which kept their laws the same. This in turn means that it is possible to test the theory that these licencing laws hurt consumer welfare by raising agent earnings and lowering agent competition. By doing so this test will potentially support the findings from the random effects model regressions in phase 1.

3.2 The Models

The models for the first phase of regressions take the general form:

𝑌𝑗𝑡 = 𝛽0+ 𝛽1𝐸𝑗+ 𝛽2𝐶𝑗𝑡+ 𝛽3𝐷𝑡+ 𝜇𝑗𝑡+ 𝜖𝑗𝑡

Where Y is one of the four dependant variables in state j outlined above; E is a vector of explanatory educational variables in state j including number of hours of education required before sitting for a licence, annual training requirements or the number of years a salesperson must be licenced before they are eligible to become a broker; C is a vector of control variables in state j at time t, including percentage of urbanisation house prices state GDP growth and housing sales amongst others; D is a year dummy variable; μ represents unknown parameter vectors; and ε is an i.i.d error term.

The models for the second phase of regressions take the general form:

𝑌𝑗𝑡= 𝛽0+ 𝛽1𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑗+ 𝛽2𝐴𝑓𝑡𝑒𝑟𝑟𝑒𝑓𝑜𝑟𝑚𝑡+ 𝛽3(𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 ∗ 𝑎𝑓𝑡𝑒𝑟𝑟𝑒𝑓𝑜𝑟𝑚)𝑗𝑡+ 𝜀𝑗𝑡

Where treatment is a dummy variable equal to 1 if the state is in the treatment group and 0 otherwise; afterreform is a dummy variable equal to 1 if the year is after 2011 when the reforms where enacted; and treatment*afterreform is the interaction term between the two dummies which shows the effect of the treatment controlling for the general trend and ε is an i.i.d error term.

(25)

3.3 The Data

The data used in this thesis has all been collected from official or industry sources. The remainder of this chapter will cover what data has been collected, the sources involved, how the data has been manipulated before being used in the regressions, what gaps exist in the available datasets and any implications that this may have for the accuracy of the models. Panel data, collected from the 50 U.S states over the periods 2007-2014 forms the basis of analysis. There are 19 variables ranging from earnings data to house prices. These variables are listed in Table 3.1below

Table 3.1 Variable Definition

Variable Name Definition Dependant

b_earnings Broker annual earnings by state

s_earnings Salesperson annual earnings by state

no_B Number of Brokers per 1000 population in the State

no_S Number of Salespersons per 1000 population in the State

Explanatory

s_eduhrs Salesperson educational requirement before

sitting for licence

s_CPD Salesperson Continuing Professional

Development annual hours

b_eduhrs Broker educational requirement before sitting for broker licence

b_CPD Broker Continuing Professional Development

annual hours

yrs_sexp How many years an individual must be licenced as a salesperson before becoming a broker

Control

(26)

Pcent_SGDPchange Percentage change in State GDP since last year Pcent_urbanpop Percentage of urban population within the state

House_price Average house price in state

state_earn Average State individual earnings

Existing_Housing How many existing homes were sold in the state that year

New_Construction How many new homes were sold in the state that year

Unemploy State Annual unemployment rate

The salesperson and broker licencing requirements include any post licensure requirements imposed by the state body. Post licensure educational requirements are additional hours of coursework that an agent must complete, after they pass the state exam, typically before their first licence renewal period ends. It forms part of the educational requirements required before an agent is considered fully qualified and for that reason the two totals have been combined. Continuing Professional Development (CPD) is ongoing ‘refresher’ training which agents must complete each licence period which varies between states. For the purposes of direct comparison it has been converted into an annual figure. Many states also require that a prospective Broker spend a number of years as a salesperson before they are eligible to sit the broker’s exam – yrs_sexp contains this data.

In general it is expected that educational hours and years of experience required will have a positive effect on earnings but a negative effect on numbers as they raise the bar for entering the profession. Conversely CPD requirements would be expected to have a negative effect on both earnings and numbers as they impose an annual cost on agents, and discourage inactive agents from remaining current.

The control variables are fairly self-explanatory. They cover a range of economic and real estate specific factors which could potentially influence agent earnings or numbers. They are included in the regressions to minimise any omitted variable bias in the model. The regression result tables in chapter 4 will show the regressions in a basic

(27)

form with only year dummies; with wealth only controls as well as the full model will all of the control variables listed above. This will add insight into the overall strength of the model by showing how the coefficients for the explanatory variables change as the model is expanded.

Data on the numbers of agents and educational requirements comes from the Association of Real Estate Licence Law Officials (ARELLO) an industry body which conducts an annual survey of state real estate regulatory bodies and maintains a database of real estate licence statistics. The U.S Bureau of Labor Statistics (BLS) which conducts an occupational wage survey provides agent and state wide earnings data as well as the annual state level unemployment rate data. The U.S Bureau of Economic Analysis (BEA) provides state level economic data such as GDP growth. Census data and projections from the U.S Census Bureau are used for urbanisation percentages and calculating per capita figures. All data has been converted in to constant 2014 dollars using the BEA’s published inflation figures.

3.3.1 Summary Statistics

This section will provide a summary of the data being used in this paper’s regressions. It gives reader an idea of the shape of the data and the overall state of play of key variables before any regression analysis is performed. Summary statistics for dependant variable data is shown in Error! Reference source not found. below

Table 3.2 Dependant Variable Summary

stats Salesperson Earnings Broker Earnings # salespersons # brokers mean 54142.09 77003.96 3.678831 1.683148 std. dev 12286.88 21996.14 2.270296 1.388223 min 32744.54 29313.95 0.1907953 0.367783 median 52548.83 73905.76 3.062345 1.335193 max 104003.4 151705.6 11.70084 7.938395

(28)

Table 3.2 shows a summary of the four dependant Y variables being tested in this paper. It indicates that on average there are more salespeople than brokers in a given state, and that brokers typically earn more. This is consistent with a two tier licencing regime. Also of note is the size of the variance and range within the data, particularly in numbers of salespersons and brokers per 1000 population.

Table 3.3 Explanatory Variable Summary

stats s_eduhrs s_CPD b_eduhrs b_CPD yrs_sexp mean 92.3913 11.40217 109.8 11.85 2.2

sd 43.98042 5.521688 100.9227 5.286764 1.021081

min 24 4 8 4 0

median 90 10.25 90 12 2

max 270 30 630 30 5

Table 3.3 summaries the educational requirements to become a real estate agent which serve as the explanatory variables in this paper. It shows that there is significant variation in all 5 variables across different states.

Table 3.4 Control Variable Summary

stats Per Capita State GDP Annual % Change GDP % Urban Pop Ave House Price State Earnings Existing Housing Sales New Construction unemp mean 46892.83 0.97959 73.59 260627 45056.54 91550 8747.493 6.8425 sd 8745.797 2.65293 14.43684 124216.6 5386.78 103429.4 12117.35 2.208322 min 31243 -9..238 38.7 139581.7 36011.29 7667 401 2.6 median 45711 1.0669 73.75 231862.6 43729.9 66619 5176.5 6.7

(29)

max 71047 19.50753 95 979968.9 58300.27 614316 94278 13.6

Lastly Table 3.4 above outlines the 8 control variables used in the regressions. These variables track the U.S economy over the period in question. The wide spread of numbers within each variable attest to the diverse effects of the great recession in the U.S, particularly in each state’s housing market. By controlling for these variables and including year dummies the regressions outlined above will minimise the potential for omitted variable bias in the results.

3.3.2 Data Shortfalls

This section will briefly discuss some of the shortcomings and limitations of the available data with the aim of placing the utility of the data in context before discussing the regression results in the next chapter.

There are two principle issues with the dataset being used in this paper. The first is that some key variables are missing values for some states or some years. This mostly affects the dependant variables for salesperson and broker numbers. This gap occurs when the state regulatory body fails to complete this question in ARELLO’s annual survey. In some cases this data has been available from the state agency website, however there are still gaps. These gaps will increase the standard errors of the regressions and lower the likelihood of finding significant results.

The second shortfall of the data is that not all variables vary over time. This limitation has necessitated the move from a fixed effects to a random effects regression model, potentially creating an omitted variable bias problem. However as discussed in the methodology section above, this possibility is being minimised through the use of control and year dummy variables as well as a separate difference in differences regression to test the findings.

(30)

4 R

ESULTS

This chapter will present the results of the regressions conducted in accordance with the methodology outlined in chapter 3. It will first present each of the earnings model regressions in turn followed by the number of agents’ model regression. It will then discuss how these results compare with economic theory as well as past papers. The second section will present the results of the second phase case study regressions involving a difference in difference analysis of the effect of Illinois’s increase in education requirements for both real estate agents. This section will further discuss how the results support and confirm the results from phase 1. Detailed analysis of the results will be covered in Chapter 5.

4.1 Random Effects Model Results

The two random effects regressions form the core of this thesis’s analysis of occupational licencing and answering of the research question.

4.1.1 Agent Earnings

Table 4.1 below shows the output from a number of different regressions of agent earnings on education requirements and a variety of controls. It presents three versions of each regression, each with more controls in order to show how the explanatory variables respond to additional controls.

(31)

Table 4.1 Agent Earnings Regression Results (1) (2) (3) (4) (5) (6) VARIABLES Salesperson Earnings(No Controls) Salesperson Earnings(Wealth Controls) Salesperson Earnings(Full Model) Broker Earnings(No Controls) Broker Earnings(Wealth Controls) Broker Earnings(Full Model)

Salesperson educational requirement 32.27 34.16 43.46

(65.91) (48.29) (60.06)

0.624 0.479 0.469

Salesperson CPD annual hours -53.93 -83.79 -137.2

(245.0) (196.1) (211.5)

0.826 0.669 0.516

Broker educational requirement 36.33** 23.44** 19.83

(16.05) (10.58) (20.50)

0.0236 0.0267 0.333

Broker CPD annual hours 201.3 -13.19 -10.32

(328.7) (313.3) (321.5)

0.540 0.966 0.974

years of sales experience required to become a broker 3,283 594.1 921.5

(2,025) (1,752) (1,747)

0.105 0.735 0.598

Per Capita State GDP 0.138 0.287 -0.00246 0.0628

(0.230) (0.284) (0.228) (0.283)

0.547 0.312 0.991 0.824

% of Urban Population -102.8 -98.88 445.7*** 414.3**

(157.6) (164.4) (172.9) (190.1)

0.514 0.547 0.00994 0.0293

Average house price in state 0.0440*** 0.0476*** 0.00257 0.00751

(0.00979) (0.0103) (0.0145) (0.0173)

(32)

(1) (2) (3) (4) (5) (6) VARIABLES Salesperson Earnings(No Controls) Salesperson Earnings(Wealth Controls) Salesperson Earnings(Full Model) Broker Earnings(No Controls) Broker Earnings(Wealth Controls) Broker Earnings(Full Model) (0.536) (0.602) (0.591) (0.684) 0.226 0.535 0.0286 0.115 % GDP Growth -460.22** -75.40 (198.08) (421.62) 0.0202 0.858

Existing Home Sales 0.00215 0.0276

(0.0247) (0.0431)

0.931 0.522

New Home Sales -0.0743 -0.161

(0.131) (0.228) 0.572 0.480 Unemployment 168.1 52.55 (742.4) (1,164) 0.821 0.964 Constant 50,372*** 10,759 14,843 58,759*** -21,569 -15,571 (5,610) (12,761) (14,138) (6,263) (17,503) (18,461) 0 0.399 0.294 0 0.218 0.399 Observations 362 362 362 326 326 326 Number of state1 46 46 46 49 49 49 r2overall 0.0459 0.357 0.361 0.0866 0.391 0.395 Wald chi2 38.62 164.2 233.4 19.17 132.1 175.5 Prob >chi2 0.0135© 0 0 0.0238 0 0

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

P Values in Italics

(33)

The results from table 4.1 above show that neither educational requirements nor CPD hours are significantly associated with increases in agent earnings once the full range of control variables is considered. Nevertheless the signs on each of the explanatory variables are as expected with educational hours and years of experience being positive and CPD being negative in most regressions. This suggests that despite the lack of significance there is potentially still an effect on earnings. This is supported regressions (4) and (5) in the table were broker educational requirements are significant at the 5% level, although this falls away when additional controls are imposed.

The significant and positive coefficient of average house prices on salesperson earnings makes sense given that most agents work on a commission basis, as is the positive if not always significant sign on state earnings. Agents in richer states with more expensive houses are more likely to earn more. The negative sign against GDP growth could potentially be a sign of the effects particular timeframe of this paper’s analysis. The 2007 recession in the U.S saw many owners foreclosing or being otherwise forced to sell. A state that’s economy is still growing is less likely to see these forced sales, lowering agent earnings.

An interesting outcome from the control variables is the opposite signs for salesperson and brokers from the effect of a state’s level of urbanisation. Brokers see a significantly positive increase in earnings from increased urbanisation while salespeople see a drop in earnings. This suggests that urbanisation affects the two categories of agent differently. Brokers, who run whole real estate offices benefit from increased concentration of potential customers while salespersons face increased competition from other agents. The significantly positive coefficient for urbanisation in regressions 2 and 3 in Table 4.2 support this analysis.

While the majority of variables in table 4.1 do not appear to be significant, the Wald 𝑐ℎ𝑖2 and assorted 𝑟2 measurements indicate that the power of the regression increases as additional control variables are added. This suggests that any potential omitted variable bias is being minimised in the full models (3) and (6).

(34)

4.1.2 Agent Numbers per Capita

Table 4.2 below shows the output from the regression of agent numbers per capita on education requirements and a variety of controls.

(35)

Table 4.2 Numbers of Agents Regression Results

(1) (2) (3) (4) (5) (6)

VARIABLES Salesperson Numbers(no controls)

Salesperson Numbers (Wealth Controls) Salesperson Numbers (Full Model) Broker Numbers (No Controls) Broker Numbers (Wealth Controls) Broker Numbers (Full Model)

Salesperson educational requirement -0.000718 -0.00783* -0.0133***

(0.00397) (0.00420) (0.00438)

0.856 0.0623 0.00235

Salesperson CPD annual hours 0.0353 0.0154 0.0333

(0.0656) (0.0347) (0.0337)

0.591 0.658 0.323

Broker educational requirement 0.00134 0.00113 0.00234

(0.00136) (0.00117) (0.00227)

0.324 0.331 0.302

Broker CPD annual hours -0.0456* -0.0547** -0.0613**

(0.0265) (0.0257) (0.0293)

0.0857 0.0332 0.0364

years of sales experience required to

become a broker -0.631** -0.738*** -0.800***

(0.293) (0.272) (0.296)

0.0315 0.00671 0.00693

0.277 0.0168 0.905 0.750

% of Urban Population 0.0980*** 0.113*** 0.0136 0.0169

(0.0284) (0.0226) (0.0120) (0.0151)

(36)

(1) (2) (3) (4) (5) (6)

VARIABLES Salesperson Numbers(no controls)

Salesperson Numbers (Wealth Controls) Salesperson Numbers (Full Model) Broker Numbers (No Controls) Broker Numbers (Wealth Controls) Broker Numbers (Full Model) 0.0998 0.239 0.108 0.313

0.565 0.577 0.415 0.452

% GDP Growth 0.05448** -0.00487

(0.02591) (0.01153)

0.0355 0.673

0.335 0.338

(0.0313©) (0.00873©) 0.0585 0.405 Unemployment -0.0113 0.0217 (0.0914) (0.0392) 0.901 0.580 Constant 2.938*** -1.780 -3.355* 3.310*** 0.958 0.499 (0.684) (2.159) (1.822) (0.781) (1.373) (1.500) 0.0173© 0.410 0.0655 0.0224© 0.485 0.739 Observations 258 258 258 280 280 280 Number of state1 44 44 44 48 48 48 r2overall 0.0514 0.572 0.702 0.324 0.486 0.469 Wald chi2 59.23 172.2 289.4 27.78 43.26 51.02 Prob >chi2 0 0 0 0.00104 0.0780© 0.0529©

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

(37)

The results in Table 4.2 above show more significant results than those in Table 4.1 suggesting that numbers of agents per capita are more effected by educational requirements than earnings is. Firstly the full model (3) shows that an increase in salesperson education requirements is negatively associated with the number of salespeople in the state. Using the mean number of salespeople as a guide the results indicate that an increase of 10 hours in pre-licensure courses would be expected to lower the number of salespeople by 3% significant at the 1% level.

On the brokers side the results show that both CPD and experience requirements are statistically significant and negative, implying that they reduce the number of brokers in a state. The years of experience coefficient in particular at -0.8 is sizeable given that the mean number of brokers per 1000 is only 1.68. This indicates that increases in the years of experience before becoming a broker is an effective way of controlling broker numbers. Many states require a set number of transactions to be completed for a year of experience to count. Given that many real estate sales positions have very low base wages and rely on commission for the majority of earnings; these requirements likely weeds out many individuals preventing ‘uncommitted’ salespeople from joining the more lucrative broker side of the profession. While broker educational hours do not appear to be significant, this is not surprising given that the additional requirements are generally lower than for salespeople, that brokers are already involved in the profession and that the years of experience restriction plays a much more important role in discouraging new brokers.

As was noted in the discussion of table 4.1, increased urbanisation is positively significantly associated with increased numbers of agents, but neither state earnings nor average house prices appear to have much of an effect on agent numbers. Per capita GDP is significant in regression (3) but the effect is very small. As in table 4.1 the increases in the Wald 𝑐ℎ𝑖2_{and assorted 𝑟}2_{measures as the model is built up suggests}

(38)

4.2 Difference in Difference Case Study Results

The results displayed in Section 4.1 above show that a number of the explanatory educational variables used in the regressions are associated with a higher level of earnings or fewer agents per capita. However as was mentioned in the methodology the random effects model being used suffers from potential omitted variable bias. For that reason a case study of the Illinois real estate licencing law changes is being used to provide additional evidence of the effects of occupational licencing. The remainder of this chapter presents the results of this case study.

4.2.1 Agent Earnings

Table 4.3 below shows the output from the difference in difference regression of agent earnings using Illinois as a treatment and both the Midwest region and the U.S as a whole as the control.

(39)

Table 4.3 Earnings Difference in Difference Regression Results

(1) (2) (3) (4)

VARIABLES Salesperson Earnings(Mid West) Salesperson Earnings(Full Model) Broker Earnings(Mid West) Broker Earnings(Full Model)

MWtreatment 660.6 -4,555 (1,434) (2,999) 0.645 0.130 MWafterreform -6,040*** -5,271* (1,400) (3,066) 0.0203© 0.0866 MWinteraction 28,229*** 2,332 (2,496) (4,620) 0 0.614 treatment -1,722 -29,658*** (6,490) (4,174) 0.791 0 afterreform -5,830*** -5,241** (918.4) (2,147) 0 0.0152 Interaction 27,329*** 25,927*** (6,794) (5,435) 0.0691© 0.00278©

Per Capita State GDP 0.325*** 0.289*** -0.198 -0.261

(0.102) (0.103) (0.174) (0.173) 0.00154 0.00535 0.256 0.132 % GDP Growth -597.12*** -413.62** -384.16 -152.55 (188.00) (190.71) (359.77) (361.64) 0.00161 0.0307 0.286 0.673 % of Urban Population 23.38 22.84 437.1*** 418.5*** (70.19) (65.90) (121.7) (114.6) 0.739 0.729 0.000378 0.000304

(40)

(1) (2) (3) (4)

VARIABLES Salesperson Earnings(Mid West) Salesperson Earnings(Full Model) Broker Earnings(Mid West) Broker Earnings(Full Model)

(0.00820) (0.00689) (0.0124) (0.0112)

0.00830 0.00143 0.357 0.651

Average State individual earnings 0.362 0.440** 1.709*** 1.897***

(0.223) (0.213) (0.364) (0.364)

Existing Home Sales 0.0105 0.0130 0.0676** 0.0634**

(0.0143) (0.0124) (0.0292) (0.0271)

0.461 0.298 0.0211 0.0198

New Home Sales -0.0574 -0.0838 -0.384 -0.339

(0.112) (0.0955) (0.241) (0.206) 0.610 0.381 0.113 0.101 Unemployment -895.9*** -720.0** -1,841*** -1,574*** (316.7) (292.9) (584.4) (536.7) 0.00491 0.0144 0.00178 0.00360 Constant 21,750*** 20,846*** -9,124 -15,500 (4,722) (4,407) (11,058) (10,432) 0.00555© 0.00313© 0.410 0.138 Observations 393 393 326 326 R-squared 0.406 0.439 0.396 0.413

P-Values in Italics

(41)

Table 4.3 shows the results of the difference in difference regressions using earnings as the dependant variable. The two important variables to examine are MWInteraction and Interaction which show the difference in difference results for Illinois compared to the Midwest group and the national control group respectively.

The results indicate a strong association between the regulatory changes and increases in agent earnings compared to the control. Models (1) and (2) show that after the new regulations went into force, salesperson earnings in Illinois rose by between $27,000 and $28,000 compared to the rest of the Mid West or country. This change is significant at the 1% level. Considering that average earnings for salespeople between 2007-2014 is $54,142 this is a substantial increase.

Similarly Model (4) shows a nearly $26,000 increase in broker earnings compared to the country as a whole following the new restrictions. This difference does not appear in the smaller more regional control however.

Taken together these results suggest that the positive signs on explanatory variables in table 4.1 are accurate despite the lack of statistical significance and that higher educational requirements leads to higher earnings. However this will be examined in greater depth in the next chapter.

4.2.2 Agent Numbers per Capita

Table 4.4 below shows the output from the difference in difference regression of agent numbers per capita using Illinois as a treatment and both the Midwest region and the U.S as a whole as the control.

(42)

Table 4.4 Numbers of Agents Difference in Difference Regression Results

(1) (2) (3) (4)

VARIABLES

Salesperson Numbers (Mid West)

Salesperson Numbers (Full

Model) Broker Numbers (Mid West) Broker Numbers (Full Model)

MWtreatment -0.729*** -0.0896 (0.247) (0.149) 0.00344 0.549 MWafterreform 0.0717 -0.00757 (0.207) (0.163) 0.729 0.963 MWinteraction -0.691*** -0.0932 (0.254) (0.195) 0.00697 0.633 treatment -1.487*** 0.450** (0.267) (0.186) 0.0000667© 0.0163 afterreform -0.430** -0.1000 (0.175) (0.157) 0.0144 0.524 Interaction 0.582** -0.524*** (0.254) (0.192) 0.0229 0.00685

(0.0207©) (0.0206©) (0.0125©) (0.0130©) 0.0125© 0.00190© 0.238 0.222 % GDP Growth -0.03077 -0.01659 0.00307 0.00836 (0.03456) (0.03412) (0.02066) (0.02083) 0.374 0.627 0.882 0.688 % of Urban Population 0.118*** 0.112*** -0.00718 -0.00865 (0.0113) (0.0105) (0.00695) (0.00656)

(43)

(1) (2) (3) (4) VARIABLES

Salesperson Numbers (Mid West)

Salesperson Numbers (Full

Model) Broker Numbers (Mid West) Broker Numbers (Full Model)

0 0 0.302 0.188

0.00640 0.000165 0.264 0.245

0.00223 0.000374 0.882 0.778

(0.0307©) (0.0284©) (0.0180©) (0.0167©) 0.00758 0.00136 0.135 0.0851 Unemployment -0.0853* -0.0559 -0.0117 -0.00604 (0.0509) (0.0483) (0.0566) (0.0549) 0.0951 0.248 0.836 0.912 Constant -4.772*** -5.679*** 0.615 0.568 (1.008) (0.875) (0.808) (0.846) 0.00360© 0 0.448 0.502 Observations 262 262 280 280 R-squared 0.657 0.658 0.151 0.153

P-Values in Italics