Regional Science and Urban Economics 41 (2011) 499–507

Contents lists available at ScienceDirect

Regional Science and Urban Economics

j ourna l homepage: www.e lsev ie r.com/ locate / regec

A question of title: Property rights and asset values

Thomas J. Miceli a, Henry J. Munneke b,⁎, C.F. Sirmans c, Geoffrey K. Turnbull d

a Department of Economics, University of Connecticut, United Statesb Department of Insurance, Legal Studies, and Real Estate, University of Georgia, Athens, GA 30602-6255, United Statesc Department of Risk Management/Insurance, Real Estate, and Business Law, Florida State University, United Statesd Department of Finance, University of Central Florida, United States

⁎ Corresponding author.E-mail addresses: Thomas.miceli@uconn.edu (T.J. Mic

(H.J. Munneke), sirmanscf@gmail.com (C.F. Sirmans), gt(G.K. Turnbull).

0166-0462/$ – see front matter © 2011 Elsevier B.V. Aldoi:10.1016/j.regsciurbeco.2011.03.011

a b s t r a c t

a r t i c l e i n f o

Article history:Received 16 November 2009Received in revised form 16 March 2011Accepted 20 March 2011Available online 25 March 2011

JEL classification:K11P14R14

Keywords:Land title systemProperty rightsRecording systemTorrens system

This paper examines the impact of land title systems on property values. The predominant system in the U.S.and a few other countries, the recording system, awards title to claimants over current possessors, whereasthe predominant system throughout most of the world, the registration system, awards title to the currentpossessor. The theory illustrates that the title system effect on asset value depends on property-specificfactors like the risk of a claim and title system characteristics like transactions costs. A natural experiment inCook County, Illinois, where both systems existed side-by-side from 1897 through 1996, allows a test of thetheory. The evidence for commercial and industrial properties reveals that parcels tend to self-select into thetwo systems as expected with registration enhancing value once the self-selection effects are removed.

eli), hmunneke@terry.uga.eduurnbull@bus.ucf.edu

l rights reserved.

© 2011 Elsevier B.V. All rights reserved.

1. Introduction

Well-defined private property rights are fundamental to marketefficiency. As important, though often overlooked as an arcane detailby economists, is the question of title—the rule used to identify thesuperior ownership claim from among competing legitimate claims toan asset. DeSoto (2000) argues that a title system, regardless of thespecific ownership rule it maintains, is an essential feature for efficientintegration of capital and landmarkets in developing countries. At thesame time, Miceli et al. (1998, 2000) show that the specificcharacteristics of the title system can affect both the pace of urbandevelopment and the pattern of investment in real estate assets, evenin developed countries.

The theoretical analysis to date focuses on how title systemcharacteristics affect land development decisions and vacant landvalues. This paper extends the analysis to developed real estate assets,offering a stylized model showing how differences in title systemcharacteristics affect values of different types of assets. Cook County,Illinois, offers a natural experiment well-suited to empirical tests ofthe value relationships predicted by the theory. Uniquely, from 1897to 1996, Cook County operated two alternative title systems at the

same time, allowing real estate owners to select the title system fortheir property. This study uses Cook County transactions data to testthe implications of the theory for relative asset values across the twotypes of title systems.

The question of title is a general concern for a wide range of goodsother than real estate. Ownership risk from competing claims arisesfor durable assets because of imperfect records arising from theft orfraud, errors made in good faith, the high cost of verification, orincomplete or lost records. In general, there are three systems forresolving conflicting ownership claims: possession, recording, andregistration. The possession rule generally requires physical posses-sion of the asset to establish a superior claim to ownership. Baird andJackson (1984) offer the example of Twyne's Case from 1601, in whichthe seller of a herd of sheep continued to take care of the herd after itssale to a second party. The English court interpreted the uninterruptedpossession as evidence that ownership had not changed hands. A formof the possession rule survives to this day in common law countries inthe form of adverse possession for real estate, where a previouslyestablished owner can lose title by failing to retain possession (Baker,et al., 2001).

The recording system (also known as the American rule in the realestate context) awards ownership of an asset to the individual withthe earlier claim. This is the rule with which most individuals areprobably familiar in the U.S. For example, vehicle ownership goes tothe individual with the prior legitimate claim as revealed by anexamination of (public) records. Later claims arising subsequent to

500 T.J. Miceli et al. / Regional Science and Urban Economics 41 (2011) 499–507

acquisition by theft, fraud, or error typically are not honored, evenwhen subsequent purchasers acted in good faith. Similarly, prove-nance in artwork, old wines, or collectibles relies on (private) recordsto verify the chain of ownership (although, in these cases provenanceplays a dual role of reinforcing the claim of authenticity as well asproviding evidence of clear title).

The registration system, on the other hand, awards ownership tothe individual with the most recent claim.1 The English rule of the lawmerchant awards ownership of an asset to the merchant acquiring itin good faith, even if purchased from a thief; the Uniform CommercialCode applies a similar doctrine for negotiable instruments in the U.S.(Baird and Jackson, 1984, p. 300).

Real estate presents the most interesting case for comparing theregistration and recording title systems. Both title systems currentlyare used to resolve ownership questions for real estate assets,depending on the locale (Shick and Plotkin, 1978); Scotland andmost of the U.S. rely on the recording title system, while some of theU.S. andmuch of the rest of the world rely on the registration system.2

Therefore, questions about how title system characteristics affect assetmarkets are particularly relevant for real estate assets.

Our model focuses on the differences between archetypicalregistration and recording systems. The registration system assignsthe current property holder title in the event that a competing claimarises. This rule minimizes ownership risk for the current holder aswell as a new purchaser. Registration, however, requires anaffirmative action by a magistrate or land court to consummate atransaction, thereby increasing transactions costs. In contrast, therecording system awards title to the claimant over the current holder.This rule imposes greater ownership risk on current holders or newpurchasers.3 At the same time, though, the transaction can becompleted without the intercession of a magistrate or similar officialaction, and so results in lower transactions costs. The relativeadvantage of one title system over the other reflects the relativeownership risk and transaction cost effects on value and thereforevaries by property type.

Section 2 of this paper presents a stylized model of the two titlesystems. The theory predicts that asset values vary across title systemsso that, when property owners are offered the choice of systems (as inCook County), resultant asset values include an endogenous titleselection effect reflecting the relative tradeoff between ownership riskand future transactions costs for the particular property. Section 3explains the selection correction model for asset values in thisapplication. Sections 4 and 5 describe the data and report theempirical results, respectively. Section 6 concludes.

2. Theory

The two property title systems we consider, the recording systemand the registration (or Torrens) system, differ in the following way.4

1 The Torrens title system is a version of the real estate registration system thatincludes a method for compensating rejected legitimate claimants for the value oftheir lost claim. See Shick and Plotkin (1978) and Miceli et al. (1998, 2000) for furtherdetails. Not all registration systems for real estate incorporate this explicit insurancefeature.

2 Most Civil Law countries use the registration title system. Among Common Lawcountries, England, most Commonwealth countries, and Hawaii and parts of Illinois,Massachusetts, and Minnesota in the U.S. also use or have used the registration titlesystem. In many countries in sub-Sahara Africa, South America, and Asia, however,informal transactions or coercive property transfers like squatting account for amajority of new urban development over the past 15 yr (Navarro and Turnbull, 2010);these countries do not have fully functioning title systems, despite official reliance onthe registration system.

3 This risk is due to past errors or fraud in determining title. The risk is reduced, butnot eliminated, by title searches, which are conducted with each sale of the property(Baker et al., 2002).

4 For a more detailed analysis of the investment incentives effects of the twosystems on which the model in this section is based, see Miceli et al. (1998, 2000).

Under the recording system, if a successful claim ismade, the claimantreceives title to the property but must compensate the current ownerfor the value of any improvements put in place subsequently. Mostlandowners purchase private title insurance to protect them againstsuch claims.5 In contrast, under the Torrens registration systemexamined here, if a claim is made, the current owner retains title, butthe claimant is entitled to monetary compensation for the value of thelost asset, to be paid from a fund that is financed by fees assessed onthose ownerswho register their property (a form of public insurance).The fundamental difference between the two systems, then, iswhether the current owner or the claimant receives title to the landin the event of a valid competing claim.

The question of interest here is whether one of these systemsresults in a higher property value, holding all other factors constant.We consider only developed land since all properties in our sampleare developed for commercial or industrial use.6 Let Vi be the observedmarket value of a property under title system i, where i=rec, reg(recording or registration system, respectively) which consists of thepresent value of the returns from the property, R, less the presentvalue of the expected transaction costs arising from future sales ofthe property under system i, Ti, and the insurance premium (orregistration fee), πi, which reflects the title risk. Thus,

Vi = R−Ti−πi: ð1Þ

Since the title systems require compensation for either the value ofimprovements or the value of the undeveloped land, we will need todecompose the total market value as follows:

Vi = Pi + I; ð2Þ

where Pi is the value of the land under system i, and I is the value ofthe capital improvements. Note that Eqs. (1) and (2) imply

Pi = R−I−Ti−πi; ð3Þ

which reflects a competitive land market. That is, the implied landvalue equals the net expected return from the parcel, including thetitle risk. Also note that the gross returns from the property, R, and thevalue of improvements, I, are treated as being independent of the titlesystem (i.e., they are not subscripted by i). This reflects the fact that,given full and fair insurance against a title claim under both systems,the amount of capital owners of undeveloped land will invest in theirland will be independent of the title system, and will be efficient.7

Consider next the determination of the title insurance premium orregistration fee πi under the recording and registration title systems,respectively. First, under the recording system, recall that if a claim ismade, the current owner loses title but is reimbursed for the cost ofthe improvements, I. Thus, his net loss is R−Trec− I. The modelrecognizes that different properties have different ownership risks; tobegin, consider a particular property with a given risk of title claimagainst the possessor. Letting θ be the probability of a successful claim(which we assume is independent of the title system in order to

5 Mortgage companies typically require borrowers to purchase title insurance for atleast the amount of the mortgage (Miceli and Sirmans, 1995).

6 Miceli et al. (2002) consider the case of undeveloped land.7 Specifically, once a buyer has purchased a parcel of undeveloped land for Pi, he will

choose I to maximize R(I)− I−Ti−πi, where R(I) is the gross return written as afunction of I. Assuming that the owner treats Ti and πi as fixed, his optimal choice of Iwill solve R′=1, which is efficient and independent of the title system (Miceli et al.,1998). Note that there is no moral hazard problem here, despite full insurance,because the investment in the land is not lost in the event of a claim even whenpossession changes.

$ Vrec

Vreg

0 θ* 1 θ

Fig. 1. Choice between title systems as a function of title risk.

501T.J. Miceli et al. / Regional Science and Urban Economics 41 (2011) 499–507

simplify comparisons), and assuming an actuarially fair insurancepremium, we obtain for the recording system8

πrec = θ R−Trec−Ið Þ: ð4Þ

Substituting this into Eq. (1) and rearranging yield

Vrec = 1−θð Þ R−Trecð Þ + θI: ð5Þ

In contrast, under the registration system, the current ownerretains title but must reimburse a successful claimant (typically,through the public fund) for the value of the undeveloped land. Thus,the actuarially fair Torrens registration fee satisfies9

πreg = θPreg : ð6Þ

Substituting this expression into Eq. (3) and solving for Preg yield

Preg¼1

1 + θR−I−Treg

� �: ð7Þ

Finally, substituting Eq. (7) into Eq. (2) and rearranging yield

Vreg =1

1 + θR−Treg + θI

� �: ð8Þ

Consider an initial point in time during which owners are free tochoose the title system in which they wish to put their property. Eachproperty holder prefers the title system that maximizes the marketvalue of his property.10 To derive general relationships, first supposethat the transaction costs are the same under the two systems—that is,Trec=Treg=T. In this case we obtain from Eqs. (5) and (8)

Vreg−Vrec =θ2

1 + θR−I−Tð Þ; ð9Þ

which is positive given R− I−TN0. Thus, the registration systemyields the higher property value when transactions costs are the samefor the alternative title systems and all else equal. The intuitiveexplanation is as follows. Under the registration system, the cost of aclaim to the current holder is the value of the land. Thus, the existenceof title risk reduces land value like a property tax. In contrast, underthe recording system, the cost of a claim to the current holder is thelost future income stream. In this case, title risk acts like an income taxon land earnings. And since a given percentage income tax reducesvalue more than the same percentage property tax, the cost of a claimis higher under the recording system.

Now suppose that transaction costs under the two systems differ.This is important in practice because the available evidence suggeststhat the cost of completing sales transactions for registered propertiesin Cook County is considerably higher than for comparable propertiesunder the recording system, reflecting understaffing and otherbureaucratic impediments associated with the registration system(Shick and Plotkin, 1978, pp. 139–142). Thus, it seems likely that, at

8 This analysis assumes that both systems apply the same standards to evaluate thelegitimacy of claims. The relative value relationships derived in this paper extend tothe case where the title systems apply different claim standards.

9 The framework assumes that both private and public title insurance are actuariallyfair in order to maintain the same basis for comparison. Differences in the two types ofinsurance funds will, of course, be reflected in premia and will have predictable effectson asset value under the two alternative title systems. These differences would bemarket-wide and therefore are not likely to affect relative values differentially acrossindividual real estate assets.10 In the Chicago example, subsequent owners could change title systems by goingthrough a legal process and paying statutory fees. Such changes, of course, also inducethe relative title risk advantages/costs of the new system, so we expect subsequentchanges in chosen title system to reinforce the self-selection effect identified in thissection by “correcting” the choice of title system for properties that could be greatlyadvantaged by such a change.

least for our sample of properties, TregNTrec. This obviously gives anadvantage to the recording system, all else equal. In the extreme casewhere there is no risk (i.e., θ=0), both Eqs. (5) and (8) become

Vi = R−Ti: ð10Þ

In a risk-free world, all landowners would prefer the recording systemso as to minimize transaction costs. At the other extreme, if the risk ofa claim is high, say θ=1, then

Vreg−Vrec =12

R−I−Treg� �

; ð11Þ

which again is positive. Thus, when the risk of a claim is very high, theregistration system is preferred despite the higher transaction costs.11

Now observe that the asset value Vi is decreasing in θ under bothsystems. It must therefore be the case that there is critical value of θ,denoted θ*, such that the recording system is preferred for θbθ* (lowrisk properties) whereas the registration system is preferred for θNθ*(high risk properties). Since different properties have different θvalues, this self-selection effect can be shown graphically as in Fig. 1:Given a set of otherwise identical properties with different title risks,Vrec lies above Vreg for low risk properties (properties with θbθ*) andVrec lies below Vreg for high risk properties (properties with θNθ*). Insummary, given the transactions cost differences across title systems,the underlying risk of claim determines whether a particular propertyhas greater value in the recording or registration system, which inturn determines the system under which the property most likely willfall.

3. Empirical model

The theoretical model implies that, when given a choice, propertyowners will find it beneficial to register their property under theTorrens registration systemwhen the title risk for the property is high,while they will find the recording system beneficial when the title riskfor the property is low. Thus, the relative value under the twoalternative title systems should play a role in an individual's choice oftitle system. The total market value of an improved property underthe registration system can be written as:

V1i = α1 + β′1X1i + u1i ð12Þ

11 This is consistent with the fact that the Torrens system was first introduced inCook County in the aftermath of the Great Chicago Fire to deal with the loss of officialland records (Shick and Plotkin, 1978, p. 139).

13 For the purpose of this study, all parcels contained within the data set are assumedto be located at the center of the quarter section, a one half mile square area, in whichthey are located. This location assumption allows for the easy computation of the

502 T.J. Miceli et al. / Regional Science and Urban Economics 41 (2011) 499–507

while the total market value of an improved property under therecording system can be written as:

V0i = α0 + β′0X0i + u0i ð13Þ

where Vsi is the natural logarithm of the ith property's selling price,and Xsi represent vectors of property and location characteristicsunder title system s. The individual's choice of title system depends onthe relative benefit of each title system choice (V1i−V0i). Althoughthe relative benefit is not directly observable, the title system choice isobserved upon the sale of the property. Thus, the underlying choice orselectivity model is:

I�i = γ V1i−V0ið Þ + θyi−ηi

I�i = γ α1 + β1X1i−α0−β0X0ið Þ− ηi + γ u0i−u1ið Þð Þð14Þ

I�i = ω′Zi−η�i ð15Þ

where Ii* is the underlying response variable (I=1 if Ii*N0, otherwiseI=0); ω is a vector of parameters to be estimated; and Zi is a vector ofregressors that determine the choice of title system (those found in thevalue equations (Xsi), as well as other factors that may determine thechoice of title system (yi)); and ηi*~N(0,σ2). Note that the error term ofthe choice equation contains the errors from the total value equations,indicating selection bias. Thus, in modeling the total value, one mustaccount for the non-random selection process underlying the choice ofa title system; failing to do sowould lead to biased estimates of the totalvalue parameters. In other words, E(usi|I=s) may not be equal to zero.

To properly analyze the impact of the title system on total value,the empirical model must be constructed in a manner that treats titlesystem choice as endogenous and also recognizes the potential forselection bias. In addition, the model must take into account that totalvalue is not jointly observable under each of the title systems. Thetotal value equation must be estimated conditional on the sample towhich it belongs. The conditional expectation of the total valueequations can be written as:

E V1i j I = 1ð Þ = α1 + β1X1i−σ1ηϕ ωZið ÞΦ ωZið Þ

� �ð16Þ

and

E V0i j I = 0ð Þ = α0 + β0X0i + σ0ηϕ ωZið Þ

1−Φ ωZið Þ� �

ð17Þ

where ϕ is the standard normal probability density function, and Φ isthe standard normal distribution function. Note that σsη measures thecovariance between the error term in regime s and the error term ofthe choice function. The significance of the parameter on σsη is a testfor selection bias. If this covariance differs from zero, selection bias isindicated. Estimating the conditional expectation of the total valueequations over the appropriate subsample provides consistentestimates of βs, but does not lead to a direct measure of the impactof title systems.

Lee et al. (1979) provide a general framework for estimating aswitching regression model with endogenous switching.12 Thisapproach is based on the estimation of an unconditional expectedtotal value equation. In addition to allowing themodel to be estimatedwith an endogenous regime choice, it also provides a direct measureof the impact of the title system choice. The unconditional expectedtotal value equation is:

E Við Þ = E V1i j Ii = 1ð ÞΦ1 + E V0i j Ii = 0ð ÞΦ0 ð18Þ

12 Heckman (1976) also considers an endogenous switching model that only allowsthe constant term to shift across regimes.

where Vi is the natural logarithm of the selling price of the ithproperty, and Φs is the probability of choosing title system s.

The theoretical model focuses on property value differences fromthe title systems that arise from title risk and transactions costdifferences. Therefore, we assume identical location and physicalcharacteristics on value across title system regimes, so that titlesystem differences show up as differences in intercepts. SubstitutingEqs. (16) and (17) into Eq. (18), constraining the slope parameters tobe constant across title system equations, and using X1i=X0i=Xi,yield the estimable asset value model

Vi = α0 + βXi + α1−α0ð ÞΦi + σ0η−σ1η

� �ϕi + λi: ð19Þ

The estimated coefficient on Ф provides a direct measure of theimpact of the title system (i.e., the difference in intercepts). Theparameter attached to ϕ provides insight into the selection process.

In addition to a direct measure of the impact of title systems, thisendogenous dummy variable approach also provides insight into theselection process. The theoretical argument indicates that a property inthe registration system should have a higher value in that system thanunder the recording system. Note that it does not indicate a relationshipbetween the absolute magnitude under each title systems (i.e.,registered properties need not yield a higher overall value). Althoughunobservable, the expected total value under the recording system,given that the property is under the registration system, is:

E V0i j I = 1ð Þ = α0 + β0X0−σ0uϕΦ

ð20Þ

while the expected price under the registration system, given that theproperty is under the recording system, is:

E V1i j I = 0ð Þ = α1 + β1X1 + σ1uϕ

1−Φ: ð21Þ

A comparison of the difference of the expected total value equationswithin group s (e.g., E(V1|I=1)−E(V0|I=1)) provides information onthe selection process andwhether the choice of title system is made ina manner that yields a higher value.

4. Data

The empirical analysis is built upon a uniquedata set drawn from theproperty transactionswithin CookCounty, Illinois. Cook County is foundin the northeastern corner of the state and fully encompasses the citylimits of Chicago. Cook County is unique in that it is the only U.S.jurisdiction in which both recording and registration title systemsexisted simultaneously, from 1897 to 1996. The Torrens registrationsystem was introduced to the area in response to the loss of nearly allland records in the Chicagofire of 1871. The cost of running two systemsprompted repeal of the smaller Torrens system in 1996.

The study makes use of transaction level data, not appraised orassessed values. The sales data were compiled by Experian (formerlyTRW-REDI) from the real estate transfer declarations, site visits, andvarious other documents. Illinois state law requires that real estatetransfer declarations be filed with each real estate sale. The sales dataprovide site and improvement characteristics, as well as informationon the location of the parcel.13 The sales data have been augmentedwith neighborhood characteristics from a second data sourceobtained from the Northeastern Illinois Planning Commission

straight line distance (in miles) for each sale to particular locations, such as the citycenter. The center of the central business district is identified as the intersection ofMadison and State Streets.

Table 1Descriptive statistics — commercial and industrial property transactions, Cook County, Illinois.

Full sample Non-Torrens (recorded) Torrens (registered)

Mean sd Mean sd Mean sd

Selling price 298,502 496,830 299,317 511,812 289,723 290,794Distance to the city center (in miles) 10.6705 6.3210 10.3276 6.1362 14.3635 7.0784Property located within Chicago city limits 0.5642 0.4960 0.5831 0.4932 0.3602 0.4814Distance to O'Hare (in miles) 12.2302 6.3209 12.3291 6.2228 11.1650 7.2284Distance to roadway activitya 3.6688 2.0614 3.6696 2.0571 3.6593 2.1125Distance to railway activitya 3.6466 2.9915 3.5318 2.9131 4.8827 3.5116Distance to open conservation spacea 2.5307 1.9672 2.5899 1.9979 1.8936 1.4583Distance to open recreational spacea 1.1926 0.6062 1.1800 0.6040 1.3276 0.6152Total building area (sq ft/10,000) 0.9756 1.7166 0.9969 1.7770 0.7462 0.7883Lot area of the parcel (sq ft) 20,101 44,107 20,236 44,883 18,647 34,745Age of improvement (years) 43.5669 26.9893 44.1528 27.2994 37.2581 22.5050Commercial land use (=1) 0.7008 0.4580 0.7014 0.4577 0.6935 0.4623Industrial land use (=1) 0.2992 0.4580 0.2986 0.4577 0.3065 0.4623Property sold in 1986 or 1987 (=1) 0.4792 0.4997 0.4838 0.4999 0.4301 0.4964Property sold in 1990 or 1991 (=1) 0.5208 0.4997 0.5162 0.4999 0.5699 0.4964Concentration of registered properties (=1)b 0.5139 0.4999 0.4893 0.5000 0.7796 0.4157Probability of Torrens registration 0.0852 0.0897 0.0756 0.0762 0.1885 0.1436Selection variable 0.1339 0.0968 0.1247 0.0906 0.2332 0.1065Sample Size 2189 2003 186

a These distance variables represent the distance in miles to the nearest quarter section with a high concentration of railway activity, roadway activity, and open space,respectively. Areas with high concentrations of a land use are arbitrarily defined as quarter sections with more than 32 acres (20%) of the land devoted to a particular land use.

b A property is defined as in a high concentration of registered properties if 5 or more registered properties exist within one mile.

14 See Brownstone and DeVany (1991), Colwell and Munneke (1997), Colwell andScheu (1994), Isakson and Ecker (2001), McMillen and McDonald (1989, 1991), andMunneke (1996).

503T.J. Miceli et al. / Regional Science and Urban Economics 41 (2011) 499–507

(NIPC). The NIPC data were used to identify quarter sections (a one-half mile by one-half mile square) having a high concentration in fourparticular land uses: railway, roadway, conservation open space, andrecreational open space. For example, quarter sections with 20% of theland area devoted to rail transportation were identified as having ahigh concentration of rail activity. (Note that in the case of railwayactively, it is not possible to distinguish between passenger andfreight lines.) The NIPC data were obtained for 1990, under theassumption that land uses (i.e., roads, parks, etc.) as a percentage of aquarter section remained sufficiently constant over the study period.The straight-line distance from each sale to the quarter sectionidentified as containing a high concentration of a particular land usewas measured.

Property index numbers, found within the sales data, helpedidentify if properties were registered using the registration or Torrenssystem. To test the accuracy of the property index number system, asample of the Torrens properties were verified using public recordsfrom the County's Registrar Office. The data were drawn from twodistinct time periods; the calendar years of 1986 and 1987 and 1990and 1991. The years between these time periods were omitted due toproblems within the data and the high cost of correcting theseproblems. The time periods used also provide an interesting backdropin that they represent a time period of normal operation, and one a fewyears prior to the 1996 repeal of the Torrens system within the state.

The full sample contains 2189 transactions of commercial andindustrial properties. Commercial (office/retail) properties make up70% of the sample. Of the full sample, only 186, or 8.5% of the sample,are identified as Torrens properties. Table 1 provides a statisticalsummary of the full sample and each sub-sample (non-Torrens,Torrens). On average, Torrens properties tend to be located furtherfrom the city center. The size of the improvements found on a Torrensproperty is smaller on average than found on non-Torrens property. Inaddition, they tend to be located on relatively larger lots.

5. Empirical results

The first stage estimates the probit model over the full sample ofcommercial and industrial properties and thenuses the probit estimatesto estimate the probability of a parcel being a registered using theregistration system, as well as to obtain an estimate of the selection

variable. The second stage estimates separate commercial and industrialvalue equations including the endogenous registration variable and theselection variable constructed from the probit estimates.

Table 2 reports the title choice equation estimates. The resultsindicate that timing characteristics play a significant role in thelikelihood of a property being in the registration system. Propertiessold in the latter time period aremore likely to be a Torrens-registeredproperty, suggesting that, as the repeal of Cook County's registrationsystem approached, properties were more likely to enter that system.Registration is also more likely as the age of a property increases,though this likelihood increases at a decreasing rate. Propertieslocated further from the city center, as well as properties locatedfurther from concentrations of recreational areas, are also more likelyto be registered. As lot size increases, the probability of registering theproperty decreases. Registration system activity by surroundingproperties increases the likelihood of a particular property using theregistration system, a result consistent with the notion that pervasivetitle problems are not necessarily isolated in individual properties, butmay also reflect underlying neighborhood-wide title problems.Although many of the individual parameters are not statisticallysignificant, we can reject the null hypothesis that all of the parametersare equal to zero at the 5% level.

The value equation estimates provide some interesting results. Thelow frequency of Torrens-registered properties in the sample requiresthat a single probit model be estimated over the entire sample.However, the sample can be split by land use type when estimatingthe value equations. Table 3 reports the commercial and industrialsample total value equation estimates. Overall, the estimated impactof the physical and site characteristics from the commercial andindustrial estimation are consistent with expectations, as well asbeing consistent between the models. The coefficients on the lot areavariables are significantly different from one and of a magnitude lessthan one, indicating that the price per unit of land is not constant withrespect to lot size, but declines as lot sizes increase. This type of valueconcavity is consistent with previous studies of land values.14 Asexpected, total value increases as building area increases. The age of

Table 2Title system choice equation.

Probit

Torrens=1

Intercept −1.7934(0.008)

Distance to city center (in miles) 0.0678(0.012)

Located outside Chicago city limits (=1) 0.3309(0.249)

Distance to city center×located outside Chicago city limits −0.0138(0.635)

Natural logarithm of lot area (sq ft) −0.1410(0.037)

Total building area (sq ft/10,000) −0.0291(0.562)

Age of improvement (years) 0.0186(0.017)

Age of improvement squared −0.0002(0.009)

Proximity to O'Hare Airport (spline)a −0.0024(0.978)

Proximity to roadway activity (spline)a −0.4808(0.165)

Proximity to railway activity (spline)a −0.0999(0.701)

Proximity to open conservation space (spline)a −0.1652(0.279)

Proximity to open recreational space (spline)a 0.6376(0.001)

Property sold in 1990 or 1991 (=1) 0.1894(0.031)

Concentration of registered properties (=1)b 0.8479(b0.0001)

Likelihood ratio 193.66(b0.001)

Note: p-values are reported in parentheses.a The proximity variables are defined as the distance to a high concentration of a

particular land use (e.g., roadway) in miles less lambda miles, and zero for distancesgreater than lambda miles away. In each case, lambda is set equal to 1 mi, with theexception of the airport variable where it is set to 5 mi. The Max-rescaled R-Squareequals 0.1920.

b A property is defined as in a high concentration of registered properties if 5 or moreregistered properties exist within one mile.

Table 3Commercial and industrial hedonic price equation.

(1)Com.

(2)Ind.

Intercept 9.0691 8.2758(37.21) (25.32)

Distance to city center (in miles) −0.0395 −0.0680(4.67) (4.93)

Distance to city center×located outside Chicago city limits 0.0155 0.0554(1.67) (3.65)

Located outside Chicago city limits (=1) 0.0044 −0.2638(0.05) (1.95)

Natural logarithm of lot area 0.3720 0.4571(13.82) (13.16)

Total building area (sq ft/10,000) 0.3582 0.1017(7.59) (5.61)

Age of improvement (years) −0.0158 −0.0191(6.37) (4.51)

Age of improvement squared 0.0001 0.0001(4.93) (3.21)

Proximity to O'Hare Airport (spline)a −0.0809 −0.0773(2.05) (3.01)

Proximity to roadway activity (spline)a 0.0589 0.0641(0.35) (0.44)

Proximity to railway activity (spline)a 0.8158 0.3950(10.5) (3.90)

Proximity to open conservation space (spline)a −0.0691 −0.0591(1.40) (0.80)

Proximity to open recreational space (spline)a −0.2157 −0.1950(3.75) (1.75)

Property sold in 1990 or 1991 (=1) 0.1596 0.2485(5.71) (5.83)

Probability of Torrens registration 1.2158 −2.9336(2.43) (3.92)

Selection variable −0.1432 3.5683(0.31) (4.90)

Adj R-sq 0.4952 0.6549

Notes: The dependent variable is the natural logarithm of sales price. The Lee et al.(1980) error adjustment is used to calculate the asymptotic t-statistics reported inparentheses.

a The proximity variables are defined as the distance to a high concentration of aparticular land use (e.g., roadway) in miles less lambda miles, and zero for distancesgreater than lambda miles away. In each case, lambda is set equal to 1 mi, with theexception of the airport variable where it is set to 5 mi.

504 T.J. Miceli et al. / Regional Science and Urban Economics 41 (2011) 499–507

the improvements is found to have a negative effect on total value, butthis effect diminishes as the building becomes progressively older.The significance of the parameter on the higher-order age variableindicates that a greater degree of flexibility is needed than can beachieved if the first-order variable is used alone.

The impact of proximity to the city center is consistent across themodels, but the local effect of the high concentration of other valuefactors tends to vary by land use. The value gradient is found to fallwith greater distance from the city center, but the gradient is lesssteep outside of the City of Chicago for both the commercial andindustrial land uses. The industrial land value gradient not onlychanges magnitude at the city boundary, it also exhibits a significantdiscontinuity. The remaining location proximity variables are mea-sured to have a local impact (i.e., are not allowed to distort the entireurban price surface).15 For both land uses, total value falls as distanceto O'Hare Airport increases. Values for both land uses increase asproximity to rail activity decreases, a result consistent with an overallocation of these types of land near rail activity. Commercial land

15 Each of these variables is constructed in a manner which measures the localimpact on price. As distance from the high concentration increases, the variableincreases until the boundary distance is reach. For distances at or greater than theboundary, the price surface is not allowed to be impacted by this land use; the variablehas the value of zero. The boundary distance of one mile was found by varying itsdefinition and finding the model of best fit.

uses are positively impacted by increased proximity to concentrationsof open space devoted to conservation and recreation. Price levels arehigher in the early 1990s relative to the mid 1980s for both land uses.

The estimates of the endogenous title system and selectionvariables are of primary interest in this study. The theoretical modelindicates that owners of properties with higher relative values underthe Torrens system would select into the Torrens registration systemand vice versa. Thus, while the price impact is important, wemust alsopay attention to the sorting between systems. For the commercialland use, the registration system variable coefficient estimate issignificantly positive. This result indicates, all else equal, that aproperty in the registration system will have a higher price than if itwere in the recording system. The estimate on the selection variable isnot found to be significantly different from zero at a 5% level.Comparing the difference in Eqs. (16) and (20) and Eqs. (17) and (21)provides insight into the selection process. These differences revealthat registered properties sell for more as registered properties thanas recorded properties (E(V1|I=1)NE(V0|I=1)). This suggests thatlandowners valued the hybrid registration-recording title system inCook County, despite the greater transactions costs associated withregistration. However, the comparison also indicates that propertiesin the recording system would sell for more if sold under theregistration system (E(V1|I=0)NE(V0|I=0)). This last result ispuzzling; it suggests that property owners in the recording systemwould have done better if they had initially registered their propertiesin the Torrens system.

Table 4Descriptive statistics (matched sample).

Torrens=0 Torrens=1 Means test

Variable (1)Mean

(2)sd

(3)Mean

(4)sd

(5)t-val

Panel A: commercial dataSelling price 241,419 324,171 249,975 273,278 0.23Lot area of parcel (sq ft) 13,979 26,910 13,587 36,349 0.10Total building area (sq ft/10,000) 0.5548 0.7017 0.5378 0.5384 0.22Age of improvements 40.2171 22.5908 43.9070 23.0388 1.30Distance to CBD (miles) 13.3762 7.5406 13.5771 6.9666 0.22Outside Chicago city limits (=1) 0.5659 0.4976 0.5581 0.4985 0.13Distance to CBD×outside Chicago city limits 9.9817 10.3582 9.9725 10.0730 0.01Distance to O'Hare Airport (miles) 12.6809 8.1315 12.2197 7.6977 0.47Proximity to O'Hare Airport (spline)a −0.1182 0.4647 −0.0597 0.2140 1.30Distance to roadway activityb 3.7063 1.8137 4.0037 2.0694 1.23Proximity to roadway activity (spline)a −0.0120 0.0754 −0.0175 0.0890 0.53Distance to railway activityb 3.9582 3.2867 4.7599 3.3816 1.93Proximity to railway activity (spline)a −0.0690 0.1978 −0.0507 0.1655 0.80Distance to open conservation spaceb 1.8287 1.4909 1.8395 1.4134 0.06Proximity to open conservation space (spline)a −0.1876 0.3413 −0.1637 0.2966 0.60Distance to open recreational spaceb 1.2840 0.6211 1.2067 0.5735 1.04Proximity to open recreational space (spline)a −0.1135 0.1926 −0.1138 0.2047 0.01Property sold in 1986 or 1987 (dummy) 0.3721 0.4852 0.3721 0.4852 0.00Property sold in 1990 or 1991 (dummy) 0.6279 0.4852 0.6279 0.4852 0.00Concentration of registered properties (=1)c 0.7132 0.4540 0.7674 0.4241 0.99Probability of Torrens registration 0.1596 0.1090 0.1834 0.1521 1.44Selection variable 0.2193 0.0965 0.2252 0.1035 0.47Number of observations 129 129

Panel B: industrial dataSelling price 281,563 203,817 379,679 311,093 1.99Lot area of parcel (sq ft) 25,148 23,104 30,097 27,852 1.03Total building area (sq ft/10,000) 1.0809 1.0114 1.2180 1.0314 0.72Age of improvements 23.7193 9.9602 22.2105 11.2975 0.76Distance to CBD (miles) 15.3021 6.0453 16.1431 7.0668 0.68Outside Chicago city limits (=1) 0.8421 0.3679 0.8246 0.3837 0.25Distance to CBD×outside Chicago city limits 14.2351 7.7882 15.1337 8.6488 0.58Distance to O'Hare Airport (miles) 9.2829 7.4979 8.7780 5.3672 0.41Proximity to O'Hare Airport (spline)a −0.5968 1.0149 −0.4848 0.9227 0.62Distance to roadway activityb 3.0378 2.1390 2.8799 2.0156 0.41Proximity to roadway activity (spline)a −0.0690 0.1573 −0.0839 0.2284 0.40Distance to railway activityb 4.1425 3.0712 5.1606 3.8059 1.57Proximity to railway activity (spline)a −0.0762 0.2176 −0.0400 0.1629 1.01Distance to open conservation spaceb 1.6726 1.3283 2.0163 1.5613 1.27Proximity to open conservation space (spline)a −0.1817 0.2974 −0.1835 0.3482 0.03Distance to open recreational spaceb 1.6195 0.6262 1.6014 0.6233 0.15Proximity to open recreational space (spline)a −0.0494 0.1175 −0.0516 0.1345 0.09Property sold in 1986 or 1987 (=1) 0.5614 0.5006 0.5614 0.5006 0.00Property sold in 1990 or 1991 (=1) 0.4386 0.5006 0.4386 0.5006 0.00Concentration of registered properties (=1)c 0.7895 0.4113 0.8070 0.3981 0.23Probability of Torrens registration 0.1949 0.1176 0.1999 0.1226 0.22Selection variable 0.2488 0.1096 0.2512 0.1118 0.12Number of observations 57 57

a The proximity variables are defined as the distance to a high concentration of a particular land use (e.g., roadway) in miles less lambda miles, and zero for distances greater thanlambda miles away. In each case, lambda is set equal to 1 mi, with the exception of the airport variable where it is set to 5 mi.

b These distance variables represent the distance in miles to the nearest quarter section with a high concentration of railway activity, roadway activity, and open space,respectively. Areas with high concentrations of a land use are arbitrarily defined as quarter sections with more than 32 acres (20%) of the land devoted to a particular land use.

c A property is defined as in a high concentration of registered properties if 5 or more registered properties exist within one mile.

505T.J. Miceli et al. / Regional Science and Urban Economics 41 (2011) 499–507

The industrial results also provide interesting insight into theselection process and its impact. The estimate on the registrationsystem variable in the industrial sample is negative and significantlydifferent from zero. While on its own this initially seems like acounterintuitive result, the selection process also must be considered.The selection results show that registered properties sell for more asregistered properties, as was the case with the commercial land usemodel. The industrial estimates also indicate that recorded propertiessell for more as recorded than as registered properties (E(V0|I=0)NE(V1|I=0)). These results show that industrial properties are allocatedbetween the title systems in a way that maximizes their value.

Identification concerns are often leveled against the type of twostage model use here. The additional variable in the probit modelcontrolling for the concentration of Torrens-registered property in the

neighborhood is included to deal with the identification issue. Even so,a robustness check of the results may be in order. For this, we look to amatching estimatormodel which has seen increasing recent use in thestudy of property markets (McMillen, 2010; Ooi et al., 2010). Thepropensity score-matching method (PSM) introduced by Rosenbaumand Rubin (1983) is a naturalfit for a robustness check of the two stagemodel given the similarity of estimation requirements. Under thisform of modeling, one assumes that the selection process is accountedfor by the observable variables. The model calls for the estimation ofconditional probabilities based on a vector of observed covariates.Within the context of this paper, the probability estimates from thefirst stage are used to represent the propensity for an observation to bea Torrens (non-Torrens) property. Dehejia and Wahba (2002) notethat in using propensity scores, numerous covariates can be controlled

506 T.J. Miceli et al. / Regional Science and Urban Economics 41 (2011) 499–507

for simultaneouslywhile stillmatching on a single number. This avoidsarbitraryweighting of several variables to generate amatched sample.

To create the matched sample, a Torrens-registered property israndomly drawn from the sample. We then select a non-Torrensproperty, within the same land use category and time period sold, thathas the minimum difference in probability to the randomly drawnTorrens-registered property. The process is repeated without replace-ment to obtain a matched sample of non-Torrens properties with thesame number of observations as the Torrens-registered sample. Table 4reports the descriptive statistics for the commercial (panel A) andindustrial (panel B) property samples constructed with this procedure.For both types of property, the mean propensity score between thesamples is not found to be statistically different. It appears that theTorrens-registered properties were not matched with non-Torrensproperties with vastly difference probabilities, something that canhappen when matching without replacement. In fact, the covariates inthe sub-samples appear balanced for both types of property; only theselling price in the industrial sample exhibits statistically differentmeans for the Torrens-registered and non-Torrens samples.

Table 5 reports the hedonic price function estimates for thematched sample. Most important for our concern, the estimatesprovide additional evidence of the Torrens registration system effectson value. The Torrens variable coefficient is positive and significant inboth the commercial and industrial property models. Nonetheless,these estimates also indicate robust results for other variables whencompared with our earlier results. Building size and lot area both havepositive and significant impacts on price. Commercial property prices

Table 5Commercial and industrial hedonic price equation (matched sample).

(1)Com.

(2)Ind.

Intercept 9.1743 8.6525(16.19) (12.68)

Distance to city center (in miles) −0.0511 −0.0573(2.37) (1.57)

Distance to city center×located outside Chicago city limits 0.0495 0.0565(2.17) (1.49)

Located outside Chicago city limits (=1) −0.4660 −0.2560(2.02) (0.84)

Natural logarithm of lot area (sq ft) 0.3473 0.3770(5.90) (5.50)

Total building area (sq ft/10,000) 0.4274 0.2579(6.45) (4.72)

Age of improvement (years) −0.0085 −0.0049(1.41) (0.39)

Age of improvement squared 6.2E−05 4.4E−05(1.03) (0.21)

Proximity to O'Hare Airport (spline)a −0.1205 −0.0240(1.26) (0.56)

Proximity to roadway activity (spline)a −0.6442 −0.0444(1.53) (0.20)

Proximity to railway activity (spline)a 0.7278 0.5549(3.91) (2.71)

Proximity to open conservation space (spline)a −0.2434 −0.1900(2.14) (1.37)

Proximity to open recreational space (spline)a −0.0472 −0.0699(0.28) (0.22)

Property sold in 1990 or 1991 (=1) 0.1954 0.1721(2.57) (2.17)

Torrens registration (=1) 0.1484 0.1641(2.24) (2.28)

Adj R-Sq 0.5408 0.7023

Notes: The dependent variable is the natural logarithm of sales price. The t-statistics arereported in parentheses.

a The proximity variables are defined as the distance to a high concentration of aparticular land use (e.g., roadway) in miles less lambda miles, and zero for distancesgreater than lambda miles away. In each case, lambda is set equal to 1 mi, with theexception of the airport variable where it is set to 5 mi.

decline with greater distance to the CBD, with a significant downwardshift the function and a smaller gradient outside the Chicago citylimits. Proximity to high concentrations of railway activity signifi-cantly affects prices in both the commercial and industrial models;prices increase with distance from areas with high concentrations ofrailway activity. And commercial land uses also exhibit lower priceswith greater distance from designated conservation open spaces.

6. Conclusions

The title system represents the rule used to resolve competingclaims to ownership of an asset and, as such, is essential forimplementing private property rights. This paper analyzed twoarchetypical systems for real estate, registration and recording titlesystems, using a stylized model to show how differences in titlesystem characteristics affect values of different types of real estateassets. The registration system minimizes ownership risk for thecurrent holder but typically entails higher transactions costs. Incontrast, the recording system imposes greater ownership risk oncurrent holders but results in lower transactions costs. The advantageof one title system over the other reflects the relative ownership riskand transaction cost effects on value and therefore varies by propertytype and land parcel.

Cook County, Illinois, had both registration and recording titlesystems operating side-by-side for over a century. Cook Countytherefore presents a unique opportunity to empirically study how thetitle system affects property value in light of the theory offered here.The empirical study used commercial and industrial propertytransaction data. The theoretical model suggests that owners ofproperties with higher relative values under the Torrens registrationsystem would select into the Torrens registration system and ownersof properties with higher relative values under the recording systemwould select into that system in Cook County. Thus, the empiricalestimates of the price impact of title systemmust also pay attention tothe sorting between systems. The selection correction empiricalmodel yields estimates generally consistent with the systematicsorting implied by theory. Industrial properties that are predicted tobe more valuable in the registration system than in the recordingsystem tend to sort that way, and vice versa. Commercial properties inthe Torrens registration system have greater value in that system thanin the recording system. Commercial properties in the recordingsystem, however, do not exhibit the same consistent results.

As a robustness check, the Torrens registration system effect onproperty values was re-estimated using a propensity scoring method(PSM) to construct a match sample for Torrens properties. Thismethod provides an alternative to the selection correctionmodel usedto obtain the first set of estimates while avoiding identification andrelated econometric issues. Nonetheless, the PSM results yield thesame conclusions as the selection correction results; the Torrensregistration system effect, in particular, is shown to increase propertyvalue as indicated by the selection-corrected estimates found usingthe earlier approach.

References

Baird, D., Jackson, T., 1984. Information, uncertainty, and the transfer of property.Journal of Legal Studies 13, 299–320.

Baker, M., Miceli, T.J., Sirmans, C.F., Turnbull, G.K., 2001. Property rights by squatting:land ownership risk and adverse possession statutes. Land Economics 77, 360–370.

Baker, M., Miceli, T.J., Sirmans, C.F., Turnbull, G.K., 2002. Optimal title search. Journal ofLegal Studies 31, 139–158.

Brownstone, D., DeVany, A., 1991. Zoning, returns to scale, and the value ofundeveloped land. The Review of Economics and Statistics 63, 699–704.

Colwell, P., Munneke, H.J., 1997. The structure of urban land prices. Journal of UrbanEconomics 41, 321–336.

Colwell, P.F., Scheu, T., 1994. A history of site valuation rules. Journal of Real EstateResearch 9, 353–368.

Dehejia, R.H., Wahba, S., 2002. Propensity score-matching methods for nonexperi-mental causal studies. The Review of Economics and Statistics 84, 151–161.

507T.J. Miceli et al. / Regional Science and Urban Economics 41 (2011) 499–507

DeSoto, H., 2000. The Mystery of Capital: Why Capitalism Triumphs in the West andFails Everywhere Else. Basic Books, New York.

Heckman, J.J., 1976. Simultaneous equations models with continuous and discreteendogenous variables and structural shifts. In: Goldfeld, Quandt (Eds.), Studies inNon-linear Estimation. Mass: Ballanger Publishing, Cambridge, Mass.

Isakson, H.R., Ecker, M.D., 2001. An analysis of the influence of location in themarket forundeveloped urban fringe land. Land Economics 77, 30–41.

Lee, L.F., Maddala, G.S., Trost, R.P., 1979. Testing for structural change by D-Methods inswitching simultaneous equation models. Proceedings of the American StatisticalAssociation (Business and Economics Section), pp. 461–466.

Lee, L.F., Maddala, G.S., Trost, R.P., 1980. Asymptotic covariance matrices of two-stageprobit and two-stage tobit methods for simultaneous equations models withselectivity. Econometrica 48, 491–503.

McMillen, D.P., 2010. Price Indices Across the Distribution of Sales Prices: A MatchingApproach, Working Paper.

McMillen, D.P., McDonald, J.F., 1989. Selectivity in urban land value functions. LandEconomics 65, 342–351.

McMillen, D.P., McDonald, J.F., 1991. Urban land value functions with endogenouszoning. Journal of Urban Economics 29, 14–27.

Miceli, T., Sirmans, C.F., 1995. The economics of land transfer and title insurance. Journalof Real Estate Finance and Economics 10, 81–88.

Miceli, T.J., Sirmans, C.F., Turnbull, G.K., 1998. Title assurance and incentives for efficientland use. European Journal of Law and Economics 6, 305–323.

Miceli, T.J., Sirmans, C.F., Turnbull, G.K., 2000. The dynamic effects of land title systems.Journal of Urban Economics 47, 370–389.

Miceli, T.J., Munneke, H.J., Sirmans, C.F., Turnbull, G.K., 2002. Title systems and landvalues. Journal of Law and Economics 45, 565–582.

Munneke, H.J., 1996. Redevelopment decisions for commercial and industrialproperties. Journal of Urban Economics 39, 229–253.

Navarro, I.A., Turnbull, G.K., 2010. Squatter Settlements and Urban Development. URAGWorking Paper No. 08-06R. Georgia State University.

Ooi, J.T.L., Sirmans, C.F., Turnbull, G.K., 2010. “The Option Value of Vacant Land” URAGWorking Paper No. 06-05. Georgia State University.

Rosenbaum, P.R., Rubin, D.B., 1983. The central role of the propensity score inobservational studies for causal effects. Biometrika 70, 41–55.

Shick, B., Plotkin, I., 1978. Torrens in the United States: A Legal and Economic Historyand Analysis of American Land Registration Systems. Lexington Books, Lexington,Mass.