Intrinsic Cycles of Land Price: A Simple Model∗

Charles Ka Yui Leung Department of Economics, Chinese University of Hong Kong

Nan-Kuang Chen Department of Economics, National Taiwan University This version: April, 2005

Acknowledgement: The authors are grateful to Robert Edelstein, Nobuhiro Kiyotaki, Chun Wah Liu, Rachel Ngai, and especially anonymous referees, seminar participants of the Hong Kong Economic Assocation Meeting (2002), Chinese University of Hong Kong, Lingnan University, University of California, Berkeley for helpful comments and suggestions; Roger Kwok for excellent research assistance; the Chinese University of Hong Kong Direct Grant, Hong Kong RGC Earmark Grant, Fulbright Foundation and National Science Council of Taiwan (92-2415-H-002-007) for financial support. Part of the research is conducted when the first author is visiting the Institute of Economics, Academia Sinica, whose hospitality is gracefully acknowledged. The usual disclaimer applies. † Leung: Department of Economics, Chinese University of Hong Kong, Shatin, Hong Kong; Fisher Center for Real Estate & Urban Economics, F602 Haas School of Business, UC Berkeley, CA 94720-6105; Phone: (852)-2609-7049, (852)-2609-7158; Fax: 2603-5805; E-mail: charlesl@cuhk.edu.hk. Chen: Department of Economics, National Taiwan University, 21 Shuchow Road, Taipei 10021, Taiwan. Phone: 886-2-2351-9641 ext. 471, Fax: 886-2-23215704, E-mail: nankuang@ccms.ntu.edu.tw.

1

Abstract The cyclicality and volatility of property prices have been extensively documented. Many explanations have been proposed. This paper builds a simple dynamic general equilibrium model in which these often cited channels are assumed away. Instead, the role of intertemporal elasticity of substitution is highlighted. In this model, the land price can exhibit price cycles. Moreover, the land price always fluctuates more than the aggregate output. The welfare of different cohorts depends crucially on the land price at the period they were born. The implications of these results are discussed. Key words : land price cycle, oscillatory convergence, intertemporal substitution JEL classification : E30, G12, R20

2

1

Introduction

Cyclicality and volatility of property and land prices have been extensively documented. For instance, Borio et al. (1994) and Renaud (1997) shows that there was a global real estate cycle among the OECD countries in between 1985 to 1994. Employing the Kalman Filter and data of more than a century, Ball et al. (1996) find very significant and long period cycles in non-residential property market, and Ball and Wood (1999) find similar results for the residential property market. Ortalo-Magne and Rady (1998) show that the real housing prices in the US and UK fluctuate more than the real GDP.1 Attempts have been made to explain the boom-bust cycles of property and land prices. There is a large literature on speculation and bubbles in the housing market as well as other asset prices such as stock prices (see for example, Case and Shiller (1989), Abraham and Hendershott (1995), Sato (1995), Ito and Iwaisako (1996), Levin and Wright (1997), and Muellbauer and Murphy (1997)). Many of these studies found evidence of speculative behavior to be a significant factor causing wide swings in property or land prices. On the other hand, much research has focused on how house prices relate to market fundamentals. Stein (1995) rationalizes the appear-to-be excess house price volatility by changes in fundamentals. Given a down payment requirement and the initial distribution of household debt levels, the model demonstrates that there is a potential for multiple equilibria, and that a small change in fundamentals can generate within-equilibrium multiplier effects, leading to large, discontinuous jumps in prices. Kiyotaki and Moore (1997), Ortalo-Magne and Rady (1998), Chen (2001), among others, also generate the cyclical behavior of asset (land) prices. The crucial element of their results is that the interactions of credit constraint and collateral value (land/housing prices) generate large and persistent fluctuations in aggregate variables.2 For empirical evidence, Chinloy (1996) finds that the real estate cycles
See Leung (2004) for a review of the literature. Alternatively, property price fluctuations can also be generated by a search theoretic model, as demonstrated by Wheaton (1990).
2 1

3

depend on the lag in construction, planning, and entitlement, and also the lumpy cost of vacancy and releasing. Leung et al. (2002), Leung and Feng (2004) find that the short-term fluctuations of housing prices tend to be induced by the “down payment effect.” Based on survey data and official data, Dokko et al. (1999), Edelstein and Paul (2000), Edelstein et al. (2001) find that the fluctuations of land price can be largely explained by the change in expectations (survey data) and changes of income generated by the corresponding land. In this paper we study the dynamics of “land” prices and its welfare implications in an overlapping generations model.3 It is well known that the land price fluctuations can have important implications on the well-being of the economic agents. Historical research such as King (1973), Macfarlane (1978) find that the land market has been developed and operated for many centuries and that the distribution of the land endowment (or inheritance) can have significant implications on the intra-generational inequality as well as inter-generational mobility. Recent studies on developing country generate similar results. For instance, Singh (1982) finds that land sale in India has important implications to both intra- and intergenerational distribution of income and welfare. This paper takes a preliminary step on this direction and focuses on the inter-generational dynamics. There are several reasons to employ an overlapping generations model. When the model is interpreted literally, it helps to explain the “long cycles” identified by the empirical literature.4 Alternatively, it can be interpreted as “waves of myopic investors”.5 Moreover, overlapping generations models also generate transactions of land and real estate across cohorts naturally, which will prove to be an important mechanism of our results here. Applying an overlapping generations model to the study of the land (real estate) market is not new,6 however, this paper differs from the previous literature in the following
There is a related literature on the relationship between housing and economic growth, such as Brito and Pereira (2002). Here the focus is very different. We study the relationship between housing price fluctuation and the aggregate economy (which is “constant” in the baseline model). 4 For instance, see Ball et al. (1996), Ball and Wood (1999). 5 Among others, see Bernanke and Gertler (1989) for more discussion on this interpretation. 6 For instance, see Brueckner and Pereira (1997) and the reference therein.
3

4

Notice that we abstract from the fact that in practice. uncertainty (persistent stochastic shocks). even when the aggregate output in the model is unaffected by the fluctuations of the land price. such as the rate of time preference and the elasticity of intertemporal substitution. land prices exhibit cycles even when the aggregate output. the equilibrium land price could still display “cycles” even without these features. In fact. land lots are "discrete" and subject to the government regulations. the model is constructed in a manner that. 7 5 .aspects. and bubbles. wage and rental are constant over time.8 As an application. the welfare of different cohorts may vary and depend on the price of the land at the period they were born. See also Mountford (2002) where multiple steady states are possible without any cyclical dynamics. the path of transitional dynamics can be indeterminate. we find that the nature of the dynamics of land prices depend crucially on certain combinations of preference parameters. and even may not restore to the steady state values in some special cases. That says. Specifically. the equilibrium land price is either constant over time. the fluctuation of land price could affect the aggregate output. They include the market frictions (informational asymmetry. The land price cycles are. In particular.7 Moreover. collateral constraints. and that would strengthen the results here. 8 Empirically. which vary across countries and time periods. in a sense. we also extend the analysis to a situation See Baumol and Benhabib (1989) for an exposition of different types of dynamics. the land price in this model fluctuates more than the aggregate output. which seems to be complementary to our work. land serves as an input for production (from the aggregate perspective) and also an investment vehicle (from an individual perspective). First. government intervention (taxation and policy changes). and construction lag). even when the steady state is unique. the model is intentionally abstracted from many features which are considered in the previous studies that have been proposed to explain the cyclicality of land prices. Second. This paper complements the literature by showing that while these features are important. intrinsic. It means that the land price oscillates above and below the steady state value. or it exhibits oscillatory dynamics. We attempt to focus on the "intrinsic nature" of land price which does not depend on the particular details of the government regulations. as found in Ortalo-Magne and Rady (1998). adjustment costs. In addition.

which redistribute wealth across generations. and Gerlach and Peng (2003). especially when land played a very important in the aggregate production (Gottlieb (1976). It should be stressed that understanding the dynamics of land price can contribute to our understanding on the aggregate economy. recent empirical works confirm the prominent role of land serving as collateral in mitigating credit constraint of firms.10 Under this scenario. to name a few. the movement of land price. will have an additional impact to the economy.when the probability of second period survival depends on the first period consumption. 2000). see Kalemli-Ozcan (2002) for evidence. For example. Borio et al. when land and labor are crucial factors of production. and Sakolski (1932)). as land is an important input even for modern economies. MacFarlane (1979) documents that property market is indeed very active in England since 1400. Kiyotaki and West (2004) report that the elasticity of substitution between land and capital in the aggregate Japan production function is close to unity. Using firm-level data. In particular. or a developed economy in earlier centuries. For instance. 10 9 6 . Ogawa and Suzuki (1998. it can be shown that this extension can lead to multiple steady states and very rich dynamics in land price. IMF (2000). Macfarlane (1979). and Ogawa and Kitasaka (1999) find that the movement in land price significantly affects the investment behavior of credit-constrained firms in Japan. as well as in population size. Kwon (1998). A high land price could lead to a lower level of consumption for the young generation. (1994). Under some parameter values. Among others. which leads to a lower level of survival probability. For instance. It in turns affect the distribution of consumption in the following period. This may shed light on the widely documented historical experience of certain economies. with time series data from 1961 to 1995. and find that bank lending is closely correlated with property prices in both developed and developing economies. examine the relationship between the role of real estate prices behind the lending boom and bust during the late 1980s and 1990s. Kiyotaki and West (1996) and Ogawa and Kitasaka (1999) find that land value significantly affects aggregate investment of Japanese firms.9 This can be interpreted as the case of a developing country.

13 receives wage wt . We defend our assumption as follows. which is referred to as land thereafter. MaCurdy. it is accustomed to assume that the capital stock to be “divisible”.11 The economy lasts forever but agents only live for two periods. for instance. It will be relaxed in a later section. as the title suggests. In our view. In macroeconomic analysis. 13 Empirically. We also investigate the welfare implications land price fluctuations for different cohorts. see Hildenbrand (1974)). Our paper. but is rather created by the government regulation and “lot division”. The last section concludes. with a rate of return rt+1 at period t + 1. (1986) 14 Even though land is often considered to be indivisible. consumes c1. in the paper we take the extreme position to assume that it is perfectly divisible. Section 3 extends the analysis to non-separable utility functions. see the survey by Blundell. and purchases land lt at unit price qt .t . (1999). the total amount of consumer is also frequently modeled as a continuous variable while in reality we only have a finite number of people (for instance. attempts to focus on the intrinsic nature of the land itself and thus assumes away regulations that may vary across 11 7 . Section 4 then extends our analysis to endogenous survival probability. In microeconomics and general equilibrium theory. which is then rented out to the firms to serve as a productive input. Section 2 presents the theoretical model and illustrate the possibility of oscillatory dynamics with numerical examples. The young generation of period t supplies labor inelastically. see Becker (1993) for a review). The land will also be resold at unit price qt+1 .14 The old agent then consumes both the rental income OLG model has been widely used at least since the publication of Diamond (1965). By the same token. the labor supply elasticity for men are small. i. see Azariadis (1993).e. For more discussions.. J. for a country like U. which probably has millions of “lots” within the whole country. indivisible (for instance... the observed “indivisibility” of land is not intrinsic. For instance.. and get even smaller as the lenght of period extends.” but labor economist tend to model “ability” as a continuous variable in order to focus on more important issues (for instance. It is true in other fields as well. it is well known that fixed capital investment is lumpy. 2 A Simple Dynamic Model Consider an overlapping generations (OLG) model in which there is a fixed quantity of durable asset.S.. so that the researchers can focus on other issues. 12 Here. Despite the “indivisible nature” of the capital stock. and nothing when old.12 Each agent is endowed with one unit of labor when young. In other words. we implicitly assume that the population is constant over time to simplify the explosition. see the survey paper by Caballero (1999)). the “differentiability” can be interpreted as an approximation. Among others. R.The rest of the paper is organized as follows. measured “ability” is “discrete. Pencavel. the “discreteness of land lot” seems to be too burdensome to carry around and it seems reasonable to use the differentiable assumption instead. T.

t. 8 .t . u00 (.) < 0. Combining the equations from (3) to (5) delivers the following nations.” a transversality condition for the asset prices is imposed.) > 0. c2. and θ is the rate of time preference.t . c1.t . (3) (4) (5) To rule out “bubbles. qt is the price of land at period t. Each generation maximizes the lifetime utility as follows: µ 1 1+θ ¶ max u(c1. c2.and resale value of land before exiting form the economy.t+1 denotes the consumption of the old generation at period t + 1. This realistic restriction will help us to rule out certain equilibrium paths.) is concave.t+1 ) s.t+1 = (qt+1 + rt+1 ) lt .t ) + u(c2.t denotes the consumption of the young generation at period t.t = λ2.t+1 )/ (1 + θ) = λ2. λ2. qt λ1. Let λ1. Consumption in both periods are assumed to be non-negative.t+1 ≥ 0. (1) (2) where c1.t be the Lagrangian multipliers of the constraints (1) and (2) respectively.t + qt lt = wt .t ) = λ1. c1. u0 (. limi→∞ (1 + θ)−i qt+i = 0. The first order conditions are u0 (c1. c2. u0 (c2.t (qt+1 + rt+1 ) .t . The utility function u(. Notice that this formulation emphasizes on the production role of land.

t = w − qt . c2.expression: u0 (c1. the demand of land is equal to the supply.). At the equilibrium. The production technology is constant returns to scale in land and labor. Equation (9) can be interpreted as an asset-pricing £ ¤ equation.t+1 = (qt+1 + rt+1 ) lt . (10) where the intertemporal marginal rate of substitution Mt+j ≡ (1 + θ)−1 u0 (c2. f 00 < 0. (6) can be simplified as u0 (c1. u0 (c2. rt = r. (6) is a non-linear difference equation of qt . Implicitly. Notice that it can be re-arranged as qt = (1 + θ)−1 u0 (c2.t ) qt+1 + r ¢ . it can be expressed as a present value formula: Ã i ∞ X Y i=1 j =1 qt = r Mt+j ! . rt+1 .t+1 )/u0 (c1. together with other variables (such as wt .t+j −1 ) 9 .t+1 ) 1+θ ¡ (9) where c1. (7) where f (lt ) is the output per unit of labor given an input of lt units of land. which is normalized to unity.t = wt − qt lt .t+j )/u0 (c1. wt = f (lt ) − lt f 0 (lt ). (8) Substituting (8) into (7) shows that the return to different factors of production are constant over time. Solving recursively. = 1 0 qt u (c2. ∀t. Recall that the labor is supplied inelastically by the young generation whose population is normalized to unity.t ) qt+1 + rt+1 = . wt = w. lt = 1. it gives rise to the familiar conditions: rt = f 0 (lt ). Equipped with this.t+1 = qt+1 + r.t ) (qt+1 + r) . which describe the dynamics of the land price.t+1 )/ (1 + θ) qt (6) where c1. c2. etc. f 0 > 0. Given that the factor market is assumed to be competitive.

dqt+1 = (1 + θ)1/(1−σ) dqt µ µ ¶ ¸ ¶−1/(1−σ) ∙ w w −1 (1 − σ ) −1 . we impose further restrictions on the preference henceforth.16 Now. the land price could still fluctutate in a dynamic equilibrium model with perfect foresight agents. in which the fluctuations of land prices come from the changes in the expected future user cost as the agents’ utility function are linear. since the rental rate of land is constant under our specification. Furthermore. 15 10 . qt (12) which explicitly shows how the land price evolves over time. the “asset-specific risk” correlation between the discount factor and asset-specific rate of return is zero. then even if the rental income is constant over time. this limiting case is not considered in the main text. we assume that the utility function exhibits the typical constant intertemporal elasticity of substitution. the land price can still fluctuate. When σ = 1. c1−σ . equation (10) demonstrates the following lemma: Lemma 1 Even if the rental income of land is constant over time. σ 6= 1. Since the results in this paper applies to any positive value of σ except when σ = 1. u(c) = log c and the dynamics becomes trivial. and the probability that the value of σ being exactly unity is measure-zero. Differentiating both sides of (12) with respect to qt will deliver a key equation for the land price dynamics. σ > 0. Following the literature.15 Hence.is the pricing kernel in standard asset pricing models. −1 qt qt (13) For the consumption in the first period must be positive. the unit land price must be lower ³ ´ w than the wage. The first order derivative thus depends It stands in contrast with Kiyotaki and Moore (1997). 1−σ u(c) = (11) where the elasticity of substitution is 1/σ in this case. qt < w. It means that qt − 1 > 0. (9) can be re-written as 1/(1−σ ) qt+1 + r = qt (1 + θ) µ ¶−σ/(1−σ) w −1 . 16 See Barro and Sala-i-Martin (1995) for more discussion of the intertemporal elasticity of substitution. if Mt+j changes over time. To generate more analytical results. Clearly.

By definition. however.on whether σ is greater than or less than unity. as demonstrated in Figure 1a.17 This is not a very interesting case and our attention will switch to the following case. The second order derivative can be easily calculated to be strictly positive for any σ > 0. qt+1 → ∞. Wallace (1980) for more discussion. and if it does. q > 0. for the existence of a positive land price at the steady state. that in this system. the dynamics of the land price is unstable. the unique steady state may not exist. Therefore. Also. 11 . it is always true that (1 − σ )−1 (w/qt ) − 1 > 0. we have dqt+1 /dqt > 0. we treat the two different cases separately: 1. By (13). the equilibrium price of land will immediately jump to the steady state value and remain constant over time. 17 Among others. qt+1 → −r < 0. Notice. then −σ/(1 − σ ) < 0. since 0 < σ < 1. then −σ/(1 − σ ) > 1. As shown in figure 1. see Azariadis (1993). For exposition. It turns out that the inverse of the intertemporal elasticity of substitution σ plays a crucial role here. it may not be stable (we will verify the existence and uniqueness in our numerical implementations). that is. when an unexpected temporary shock hits this economy (such as a temporary change in taste. as in the case of monetary economy in overlapping generations models. By (13). It means that the land price could eventually be too high (or too low) and violate some equilibrium conditions. Now. Given that the second order derivative is strictly positive. the dynamical system can be either “oscillatory and stable” (1b and 1c) or “explosive” (1d). and thus (12) shows that as qt → 0. If 0 < σ < 1. −1 q q (14) However. 2. In the former. and thus (12) shows that as qt → 0. it must be that µ ¶−σ/(1−σ) w > r (1 + θ)−1/(1−σ) . qt = qt+1 ≡ q. there is only a jump variable (qt ) and no sluggish variable. σ > 1 leads to dqt+1 /dqt < 0. the price does not change at the steady state. If σ > 1. land price is always increasing.

i. even a simple model with land as a factor of production is capable of generating land price cycles under certain conditions. as the initial land price is not dictated by any of the first order conditions. we have a constant price equilibrium path. This section attempts to provide some intuition behind these results.19 Since neither of these are allowed. Furthermore. First. or even the productivity).1 Discussion of the Results The previous section shows that. 12 . It will eventually converge to the steady state value but the convergence is not monotonic. that there are infinitely many paths to the unique steady state that are also consistent with the equation (12). [Insert figure 1 here] 2. however. we encounter the situation of “path indeterminacy ” which has been studied by the macroeconomics literature.e. depending on whether the value of σ is larger than unity or not. as in the case with 0 < σ < 1. or it becomes “too low” (diverges to negative infinity) so that the second period consumption becomes negative. she is rather “passive”: she simply collects the rent and sells all her land. since labor is supplied inelastically when 18 19 See Baumol and Benhabib (1989) for more discussion. The important economic decision is made when the agent is young. the land price will either be “too high” that it will eventually exceed the first period wage income. the land price will oscillate above and below the steady state value. In other words. the price will jump to the steady state immediately.18 On the other hand. and then consumes. pushing the first period consumption to be negative. Notice. Thus. The value of land cannot be negative because land is a productive input here. notice that when an agent is old. if the system is “explosive”.in tax rate.

qt lt = (16) Notice that the rate of return for holding land Rt is increasing in qt+1 (see (15)). After some manipulations. the demand of land increases (decreases) with the (expected) future land price if 0 < σ < 1 (σ > 1). qt qt Rt ≡ (15) Given the utility functional form (11). ∂lt = ∂qt+1 µ ¶ ³ ´−2 µ 1 − σ ¶ w 1/σ (σ −1)/σ 1/σ −1/σ (1 + θ) (Rt ) . the purchase of land should be the main focus. If 13 . the task here is to characterize the young generation’s demand function for land. Consider first the case when σ is low. we define the rate of return of holding land. it can be shown that the function of demand for land by young agents is given by ´−1 w³ 1 + (1 + θ)1/σ (Rt )(σ−1)/σ . The intuition of the lemma is simple. It is thus ⎧ ⎪ ⎨ > 0 if 0 < σ < 1 . ⎪ ⎩ < 0 if σ > 1 ∂lt ∂qt+1 which proves the following lemma: Lemma 2 Other things being equal. we can rewrite the equation (6) without imposing the equilibrium condition. which means that agents are more willing to substitute current consumption for future consumption. other things being equal. 1 + (1 + θ) (Rt ) qt σ ¡ 1−σ ¢ σ Notice that all terms except clear that in the above expression are always positive. qt+1 + rt+1 qt+1 + r = . it is feasible to compute the partial derivative of the demand of land with respect to the next period land price qt+1 .young. Thus. To facilitate the discussion. Hence.

[Insert figure 2 here] 14 . when agents have a higher σ . the condition that the future land price is positive or negative as the present land price approaches zero. which leads to an explosive dynamics. |dqt+1 /dqt | > 1.they expect that the future land price will appreciate. To see why. the price will jump to the steady state and stays at that level from that period onwards. This gives an intuitive explanation why (13) exhibits a positive relationship between qt and qt+1 given a low value of σ . a higher future land price qt+1 would only suppress the current demand of land. Figure 2 provides numerical examples and graphical illustrations. It must be coupled with the conditions that the future land price is negative when the present land price approaches zero. the agent will smooth out the consumption plan and increases period t consumption given the current price of land qt . note that a higher qt+1 raises period t + 1 consumption for a given period t land holding. Notice that merely positive correlation between the present land price qt and the future land price qt+1 is not sufficient to generate unstable dynamics. While |dqt+1 /dqt | ≶ 1 depends on the combination of all of the parameter values and can only be verified numerically. On the other hand.e. and hence a decline in the current land price qt . Otherwise. then they will devote more resources to demand for land. In equilibrium. and the former leads to constant-price equilibria and the latter shows cycles. which in turn raises investment. Figure 2a and 2b draw the movement of qt+1 against qt with relatively large σ (5 and 6 respectively). since the supply of land is fixed. the agents’ willingness to intertemporal substitution also raises the current land price qt . Since the agents are perfect foresight. To maximize her lifetime utility. which means that agents are reluctant to make intertemporal substitution. can be easily calculated. and that the magnitude of the “feedback” is large enough. This also explains why in this case (13) exhibits a negative relationship between qt and qt+1 given a high value of σ . given the current price of land. This results in a decline in the demand for land.. a steady state might not even exist. i.

the life-time utility) is given by −1 Wt ≡ u(c1. the life-time utility of the period t cohort will be higher than the immediate next generation if the land price in period t is lower than that in the following 15 . it is reasonable to suspect that the welfare of agents of different cohorts could also vary over time. the welfare of different cohort will also remain constant at the equilibrium. whether the life-time utility is increasing or decreasing in qt depends on the value of σ. and that dqt+1 /dqt < 0.t ) + (1 + θ)−1 u(c2. However. the life-time utility of the period t cohort is decreasing with the price of land. At the equilibrium. Thus.t = w − qt and c2. or explosive. dWt = dqt µ σ 1−σ ¶ ⎧ ⎪ ⎨ > 0 if 0 < σ < 1 £ −σ−1 ¤ . Combining the utility function of (11) with (12). for the case of large σ (σ > 1). Now define the life-time utility of a representative agent of the cohort born at time t to be Wt ≡ u(c1. We also know that the dynamics of land price can either be oscillatory and stable.2 Welfare Implications As it is shown in the previous sub-section.t+1 ). the land price could fluctuate even with a small temporary shock (the “path indeterminacy” case). This is the case when we have an “path indeterminacy” of land price. the life-time utility of the period t cohort (or in short. w (w − qt ) ⎪ ⎩ < 0 if σ > 1 For the case of small value of σ (0 < σ < 1).2. their consumption plan is c1. This section studies the evolution of agents’ welfare in this economy. Again. On the other hand.t+1 = qt+1 + r. Since the land price oscillates above and below the steady state value.t+1 ) = µ (w − qt )−σ 1−σ ¶ w. the life-time utility of the period t cohort is increasing in the value of land.t ) + (1 + θ) u(c2. Consider the case when the equilibrium exhibits oscillatory convergence. the earlier result has already established that this corresponds to the case where the land price immediately jumps to the steady state and remain constant over time. Therefore.

are based on the infinite horizon model. To put it in another way.85. it demands a precise estimation of the parameters from the data. Then simply apply the formula that θ = β −1 − 1 yields the result.05 (for a period of 30 years). In other words. The calculation is simple. one may wonder how large the “probability” of the cycle is. Unfortunately. If the annual discount factor is 0. This section presents numerical examples to investigate the dynamics of land price implied by the model. however. The current model. and thus a widely accepted estimate of the altruism is unavailable. 2. The precise estimation of the pure life-cycle discount factor still awaits for future research. The estimate for σ is even more problematic. In an infinite horizon asset pricing model For instance. the discount factor for 20 and 30 years are 0. 262 4 to 2. the model is capable of generating oscillatory dynamics as the land price converges to the steady state value. focuses primarily on life-cycle discounting and thus the appropriate discount factor should be lower. However. how likely would these cycles occur? To answer these questions. on the other hand.442 and 0. Thus.96.period. 21 20 16 .294 respectively. see Cooley (1995). and hence the discount factor include both life-cycle discounting and altruism for offsprings. with empirically plausible parameter values. or two-period cycles where the land price always oscillates between two values.96 for annual data.4014 in our two-period OLG model in which each period corresponds to a period of 20 to 30 years.3 Numerical Examples As shown in figure 2.21 These calculations.20 It translates to a value of θ in between 1.8 (for a period of 20 years) to 130. The typical parameter value used in the macroeconomics literature on the time discount factor β ≡ (1 + θ)−1 is about 0. It should nevertheless be clear that even if the annual discount factor is adjusted to 0. the empirical literature on the existence of altruism is controversial. land price cycle is a “possibility” in this model. the implied θ would be in the range of 24.

Following this line of thought. even for quarterly data. between 0 and 100. the next section extends the analysis to different types of time-non-separable utility functions. Mehra and Prescott (1985) claims that σ is within a narrow range. In a sense. Figure 3a shows two regions of parameter combinations between σ and θ: the region delivering constant price equilibria (dark-shadowed area) and the region delivering oscillatory convergence (light-shadowed area). Kandel and Stanbaugh (1991). It puts some severe restrictions on the parameter values and might be not satisfactory to some. Due to these yet-to-be-solved problems. see the survey by Kocherlakota (1996). The issue is even more complicated here because each period of our two-period OLG model corresponds to 20 or even 30 years. It is clear that although the region which delivers the “land price cycles” is large. this paper simply experiments with a wide range of parameter values. It is not very clear how the previously mentioned estimates would apply to the current model. Abel (2002) forcefully argue that. 17 . and rational expectations asset pricing model with time-separable utility functions has been found to perform unsatisfactorily. [Insert figure 3 here] 22 For instance. θ and σ . 0 ≤ σ ≤ 10. However. The 2-period cycle cases occur on the boundary of the two regions. the range of σ is in fact much larger. it might not be very surprising because this model has thus far assumes a time-separable utility function. it is necessary (but not sufficient) to have θ > 10 and σ > 3.calibrated with quarterly data. Boldrin.22 Recent research seems to suggest that allowing for timenon-separable utility functions can significantly improve the ability for the model to match the data (Constantinides (1990). Christiano and Fisher (1997)).

it is discounted by the average consumption in the first period (“catching up with the Joneses”).t ) + 0 (17) 18 . As before.t+1 ≥ 0.t on the periodic utility in both periods.1 Habit Formation The basic structure retains in this section. c1. u(c1.t+1 . The first order conditions (4) and (5) are still valid.” In the former. the second period consumption is discounted by the amount of her own consumption in the first period (“habit formation”). Since it is empirically difficult to differentiate the “inward-looking” from the “outward-looking” preference. c1.t ) subject to the constraints (1) and (2).t ) + v(c2.t+1 . and that the consumption in both periods are nonnegative.t ) and v(c2. (3) is modified as µ 1 1+θ ¶ ∂v (c2.t u (c1. except for a change in the utility function. However.t . Under the outward-looking case. Although at the equilibrium.t . µ 1 1+θ ¶ max u(c1. 3. the individual consumption coincides with the average consumption level. this paper considers both types for completeness.t ) = λ1. The agent is aware of the effect of first period consumption c1. λ1. (1997). whereas under the latter case. ∂c1. In the terminology of Carroll et al. as demonstrated by Carroll et al.t represent the Lagrangian multipliers of the constraints (1) and (2) respectively. one is “inward-looking ” and the other is “outward-looking.t+1 . the agent fails to take into consideration that her consumption decision would have an impact to the aggregate/average consumption. (1997).t ). c1. the dynamics displayed and the welfare implication can be different. c2.3 Non-Separable Utility Function There are at least two types of time-non-separable utility function widely used in the literature. c1.t and λ2.

Carroll (2000).t+1 . In this case.t = ≡ Rt .24 In this case. To examine the dynamics more explicitly.23 It is further imposed that c2. and v (c2. the utility function takes the following form (c1.t ) . see Carroll. σ 6= 1. 0 < σ . It is clear that if c1.c1.t )−γ > 0.Combining the first order conditions and the market equilibrium conditions deliver a modified version of (6).t ) = .t . In the appendix. and Weil (1997. and the reference therein. and v (c2.t+1 = qt+1 + r.t = w − qt .t+1 . Overland. c2.t+1 − ac1. The results and the derivations are similar and therefore skipped due to the space constraint.t+1 > 0. two special cases of time-non-separable utility function are considered. σ 6= 1.t )1−σ (c2. c1.t+1 .t+1 .t+1 − ac1.t+1 (c1. then c2. u(c1. Subtractive Habit Formation This form of time-non-separable utility can be at least traced back to the seminal work of Constantinides (1990).t )−γ (c1. 1.c1.t ) = ln (c2. σ > 0. ¡ 1 ¢ ∂v(c2.t )1−σ . the utility function takes the following form ¢1−σ ¡ c2.t ) u0 (c1. and v (c2.t ) qt 1+θ ∂c2.t+1 − ac1. it is shown that equation (18) can be re-written as à µ !σ ¶−1 w Rt + a . u(c1. −1 −a = qt 1+θ Rt (19) 2. c1.t ) + 1+ qt+1 + r θ ∂c1. 2000). c1.t ) → ln (c1.t )1−σ u(c1.t ) = .t ) . c2. Multiplicative Habit Formation This form of habit formation has also been widely used in the asset pricing and macroeconomics literature.t+1 . When σ → 1. 23 19 . 24 For instance.t+1 (18) where c1. ¡ 1 ¢ ∂v(c2.t ) = . 1−σ 1−σ where 0 < γ < 1.t ) = 1−σ 1−σ where 0 < a < 1.t+1 (c1.t > 0.

subtractive habit formation is even worse than the baseline case. combining the first order conditions and the market equilibrium conditions. there is a wider range of parameters which generates cyclical land price dynamics with multiplicative habit formation. Formally. as Figure 3c shows.t .t . “Catching up with the Joneses”. has not be relaxed. Therefore.t = c1.t ). 3. qt (1 + θ) (qt )σ (w − qt )γ (1−σ)−σ 1 + γ (20) In the appendix. c1.2 Catching up with the Joneses The next class of time-non-separable utility function. there is a unique Rt that solves (19) ((20)) in the case of subtractive (multiplicative) habit formation. On the other hand. As before. c1. let λ1.In the appendix. The consumption in both periods are required to be non-negative. Figure 3b shows that. where c1. it is shown that equation (18) can be re-written as " µ ¶−1 #−1 w −1 = (Rt )1−σ .t ) + µ 1 1+θ ¶ v (c2. can be at least traced back to Abel (1990).t as given.t+1 ≥ 0. we 20 .t and λ2. The restriction on σ . Condition (3) is also valid because the agent takes the average consumption c1.t is the average consumption level of the period t. and the fact that c1.t be the Lagrangian multipliers of the constraints (1) and (2) respectively. subject to the constraints (1) and (2). it is further shown that under some conditions. The first order conditions (4) and (5) are still valid. holding other parameter constant. Virtually any value of θ can generate oscillatory convergence.t+1 . provided that a suitable value of σ is chosen. it means that the agent maximizes the life-time utility max u(c1. on the other hand. c2.

1. Equation (21) can be re-written as à µ !σ ¶−1 w Rt .t ) qt+1 + r ≡ Rt . c1.t )1−σ (c2. σ > 0.t+1 (c1. c1. It is further imposed that c2.t ) = 1−σ 1−σ where 0 < a < 1.t )−γ (c1.t+1 > 0. To compare with the habit formation case. It is clear that if c1.t )1−σ . ¢ ∂v(c2.t > 0.t ) = qt ∂c2. Subtractive Catching Up with Joneses Similar to the subtractive habit formation case.t ) = . 1−σ 1−σ where 0 < γ < 1. then we have 21 .t+1 ¡ 1 1+θ (21) which is different from both the time-separable case (see (9)) and habit formation case (see (18)).t ) = .t+1 − ac1. c2.t+1 . the utility functions take the following form: ¢1−σ ¡ c2. and σ 6= 1. 2.c1.t+1 −ac1. and σ > 0. v (c2.t ) = . we also consider both subtractive and multiplicative cases.t+1 .t )1−σ u(c1. −1 −a = qt 1+θ Rt (22) which is different from (19) by only a constant term.have a modified version of (6). u(c1.t . u0 (c1.t+1 . Multiplicative Catching Up with the Joneses Similar to the multiplicative habit formation case. v (c2. the utility functions are assumed to be: (c1.

we have assumed that the population is constant over time and all young agents will surely survive in the following period. it is shown that equation (21) can be rewritten as (1 + θ) (qt )σ (w − qt )γ (1−σ)−σ = (Rt )1−σ .25 The multiplicative case produces a wider region of parameter combinations that deliver oscillatory land prices. while the subtractive is worse than the baseline case in terms of generating cyclical dynamics. as shown in Figure 3d and 3e. See Constantinides (1990). (23) The qualitative as well as quantitative results of catching-up-with-the-Joneses preference. However. This seems to be in agreement with the empirical evidence. Li (2001) and the reference therein. as the possibility of infant death in those scenarios is not trivial. If the agent dies in the second period. see Kalemli-Ozcan (2003). The probability of survival is assumed to be positively correlated to the first period consumption. In the appendix.c2. 26 For an review of the literature. are similar to the case of habit formation. she receives zero utility and the income from land re-sale and rental will be collected by the government and then The value of a and γ used in the numerical exercises is rather small and well within the estimate of the empirical literature.t )−γ > 0. 4 Extension: Endogenous Survival Probability Thus far. In light of this. 25 22 . which is only different from (20) by one term.t+1 (c1. Campbell and Cochrane (1999). this may not be a good assumption for some historical situations.26 An agent may die due to diseases or insufficient “nutrition” she receives in the first period. or some developing countries in the modern time. among others. we relax the assumption of constant population: the second period probability of survival is no longer unity but will depend on the first period consumption.

74 (1960) to 0.re-distributed to the existing population. the agent maximizes (c1. we assume that β (c1.t . To determine the dynamics of the land price.t+1 )1−σ + β (c1.t .t ) = a0 (1 − exp (−a1 c1.t ) µ c2. τ 2. and the details of the government budget.t + qt lt ≤ wt + τ 1.82 (1997) and a1 is in between 0. β 00 (. (24) where β (c1t ) is the probability for an agent born in time t to survive in time t + 1 given an amount of consumption c1t taken in period t.44 (1997) for adult survival rate. Formally.t+1 c1. and the estimates are statistically significant.t ) ≤ 1. c1.) > 0. it is necessary to have the exact functional form of the discount factor.t+1 .t ) 1−σ 1−σ max s. Following Kalemli-Ozcan (2002). Using the World Bank data.36 (1960) to 0.t+1 ≥ c2.t ) c2. β (c1. (26) which clearly satisfies the conditions stated previously. All the income and asset of those who die 23 .t+1 ≥ 0 are the lump-sum transfer received from the government at the first and second period of time.t ) 1 − σ + (25) where Rt = (qt+1 + rt+1 )/qt . It is easy to show that 1 β (c1.t. The behavior of the government is to tax 100 percent bequests of those who die early and redistributes to the rest of the population.t )1−σ (c2. β 0 (.t ¶σ β 0 (c1. the discount factor is endogenized. τ 1. (qt+1 + rt+1 ) lt + τ 2. which is taken as given by the individual. KalemliOzcan (2002) estimates that a0 is in between 0.t )) .t+1 = Rt . Essentially. 0 ≤ β (c1.) < 0.

τ 1. however. as made clear in (24). We can then proceed with some numerical examples.t )) (qt+1 + rt+1 ) . Figure 4a presents a case with a stable steady state. in which one of them is stable (denoted P ).t )τ 2.t )) (qt+1 + rt+1 ) .27 We assume that the probability of dying is identically and independently distributed across agents. qt β 1−σ qt+1 (28) Notice that the discount factor β depends on the first period consumption c1. the government budget constraint takes the following form. τ 1. ∀t. by the Law of Large Numbers. Recall our earlier discussion that since there is no state variable and the price is a jump variable by definition. the proportion of dying in the aggregate is equal to the probability in the individual level. we assume that all of these bequests go to the surviving old. as the market clearing condition implies lt = 1.t+1 = 0.416-7).t ) − − r. In Figure 4b. Since we have assumed that the population size of each cohort is unity. (27) In the appendix. 27 24 . it is shown that the dynamics of the land price can be summarized by the following equation ∙ ¸¾1/(σ−1) ½ β0 1 σ β = (βc1. Thus.t )τ 2. It is indeed a highly non-linear difference equation. the manner of distribution will make a difference. In this case. and β (c1. the re-distribution of “accidental bequest” is not present in Kalemli-Ozcan (2002).at the end of the first period will be collected and redistributed in a lump sum manner. For simplicity. In this paper. Thus.t+1 = (1 − β (c1. Figure 4 shows that the dynamics and configuration of the steady state can be very rich. which in turn depends on period t land price qt .t .t+1 = (1 − β (c1. It remains to determine how these “accidental bequests” are to be distributed.t+1 + β (c1. we actually have three steady states. p. it is assumed that capital is not an input in the production and return to land is zero. In Kalemli-Ozcan (2002.

on top of the “convergence dynamics” towards the steady state. among others. the results may shed light on the historical experience of certain economies documented in. there will be a very dramatic changes in land price and population size (due to the survival probability dependency on the consumption of young generation). A positive death rate also means that. Consider. or experience cyclical fluctuations in land price. the economy starts with a steady state at a higher land price. In terms of dynamics. In that case. it is not easy to compare the social welfare across different steady states. For a young agent. this economy can stay at one of the two constant land price steady states. for example. While this model is highly stylized and should be interpreted with cautions. as the land price and other variables are constant over time. depends on the initial land price. And clearly. However. in ex post terms. Gottlieb (1976). the agent who survive may prefer to receive a higher payment from the government and that they can sell their land at a higher price. In other words. The young agents are then left with less income for food. On the other hand. This further discourages the young generation to save. Notice that there is a “multiple steady state” issue here and depends on the initial condition and belief of economic agents. she might prefer an economy with low land price and hence she can allocate more resource on food and increase the probability of survival. 25 . Notice also that it is not straightforward to order the welfare of agent at different steady states in a Pareto sense. with these inter-generational conflict of interest. resulting in a lower probability of survival. the “mathematically unstable steady states” are in fact very “stable” in an economic sense. other things being equal. Therefore. where the land price will oscillate around the steady state value. notice that the economy can also switch from one steady state to another when the economy is hit by a shock.an “unstable” steady state corresponds to a constant land price in this economy. and Sakolski (1932). Macfarlane (1979). the lump sum transfer from the government to the “survivors” will be higher. a “stable” steady state (mathematically speaking) corresponds to a case of path indeterminacy. the economy may rest in different steady states.

or vice versa. This paper illustrates that there is an intrinsic tendency to generate land price cycle in an overlapping generations model. The land price fluctuation would significantly affect the credit constraints of firms and hence their investment. etc. even when factors such as capital market imperfection. are abstracted away. and even wider range with multiplicative time-non-separable preference. uncertainty. it may be fruitful to introduce majority voting or other political process into the model. Furthermore. this model can potentially be extended for interesting policy analysis. land is an important factor of production even for modern economies.[Insert figure 4 here] 5 Concluding Remarks Land is a very important element in the economy. In addition. a change in some parameter or policy may move the economy from a constant-price steady state to another with land price cycles. This research can also be extended in several directions. It holds with typical time-separable utility function with a wide range of parameter values. such as an introduction of physical capital (for instance. It could translate into changes in aggregate demand and have an impact to the macroeconomy. we can have multiple steady states and very rich dynamics under certain parameter values. Thus. Clearly. this model also highlights that the interest of young agents may be very different from those who are old. non-convexity. given a large class of utility functions and a wide spectrum of parameter combinations. endogenous fertility choice (for instance. informational incompleteness. In particular. 2002). These efforts may contribute to the understanding of the cross-country difference in their historical development experience. It seems to suggest that land price cycle is inevitable in a simple general equilibrium OLG model. see Mountford. 26 . As the reference cited in the introduction suggests. we find that when the survival probability of old agents depends on the consumption of when the agents are young. especially when land played an important role in the production process.

(1995). [6] Ball. 1(4). “Structures investment and economic growth: a long-term international comparison. endogenizing the supply of land.” Housing Studies. Intertemporal Macroeconomics. Ryder and Weil. [8] Barro. New York: McGall Hill. Schlauch and Laposa (2001). 27 . 185-209. [2] Abel. Oxford: Blackwell. “Bubbles in Metropolitan Housing Markets. W. A. 40. (2002). and A. Wood (1996). R. It may be proved to be a convenient and plausible vehicle for different policy analysis. and J. Human capital: A theoretical and empirical analysis. 449-96. [9] Baumol. Morrison and A. 303-315. S. [3] Abraham. 3. 14. T. Sala-i-Martin (1995). M. with special reference to education. J. G. Benhabib (1989)." Journal of Economic Growth. J. [4] Azariadis. Shibusawa (2000)). see Chen and Leung (2003. Economic Growth.” Journal of Economic Perspectives.see Kalemli-Ozcan. 2000). and economic applications. “An exploration of the effects of pessimism and doubt on asset returns. 1993 [1964]. C. see Quah (2000).” Journal of Economic Dynamics and Control. In particular. “Chaos: significance. and P. this model can also be extended to include both intra. 26. 3rd ed. 2004). 77-105. [10] Becker. Chicago. 2003). (1993). or including residential housing (for instance.” Journal of Housing Research. 38-42. “Money and wage contracts in an optimizing model of the business cycle. Hendershott (1995). 6. 33.” American Economic Review. 28 For instance.. University of Chicago Press. (1996). 35.28 In terms of welfare analysis. M. [11] Benassy. 191-207. 1687-1706. see Wheaton 1999. and X.-P. the interaction between internet use and land use (for instance. “Asset prices under habit formation and catching up with the Joneses.. A. Leung. “Housing investment: long run international trends and volatility.H. 1075-92. allowing the land both as an input of production and also as a collateral would enable us to have a deeper understanding of the interactions of cyclical behavior in different sectors of the economy.” Urban Studies.” Journal of Monetary Economics. References [1] Abel. Wood (1999). [5] Azariadis. mechanism. [7] Ball. C.and inter-generational heterogeneity among agents. (1990). "The Economics of Poverty Traps: Part One: Complete Markets.

and Monetary Policy Implication. [19] Campbell. 79. New York and Oxford: Elsevier Science. O. P. Gertler (1989). net worth and business fluctuations. (1999). 312-332. Vol. 28 . K. “Solving consumption models with multiplicative habits. “By Force of Habit: A Consumption-Based Explanation of Aggregate Stock Market Behavior.” in Taylor. and J. 98.” Journal of Housing Research. [15] Borio. (1990). 14-31. [18] Caballero.” National Taiwan University. Amsterdam. 173-90. “The Efficiency of the Market for Single-Family Homes. [23] Chen. eds. “Habit persistence and asset returns in an exchange economy. 415-436.” American Economic Review. N. and R. [14] Boldrin. 246-71. V. Leung (2004). Cochrane (1999). Christiano and J. 3A. L. North-Holland. 339-367. “Real Estate Cycles: Theory and Empirical Evidence. “Property Markets and Public Policy Spillovers through Collateral Effect. M.” Journal of Monetary Economics. Michael. “Comparison utility in a growth model. mimeo. K.-K. Pereira (2002). 27. (2001). Card. J. R.” National Taiwan University. [24] Chen.. No. Fisher (1997). Weil (1997). Prowse (1994). “Habit formation: a resolution of the equity premium puzzle. (Eds. 67-77. D. Overland and D.40.[12] Bernanke. (1996). J. [22] Case. “Housing wealth and the economy’s adjustment to unanticipated shocks. and A. 79. Y. In: Ashenfelter.” Journal of Political Economy.” Macroeconomic Dynamics. Pereira (1997). (2000). 2. Determinants. “Bank net worth. P.” Journal of Economic Growth. Kennedy. G. and S. [26] Chinloy. and C. 48. C. 205-51. and T.). N.” Bank for International Settlements Economic Paper. 497-513. and M. [13] Blundell. 1. and A. 51. [20] Carroll.” Economics Letters. Woodford. N. 1999. R. MaCurdy.-K. (1999). “Housing and Endogenous Long-Term Growth. “Land price dynamics with endogenous supply of land. K. J. D. John. [21] Carroll. Handbook of Labor Economics. 7(2).E.-K. B. [25] Chen. Handbook of macroeconomics. 519-43. "Labor Supply: A Review of Alternative Approaches".. Shiller (1989).” American Economic Review. C. “Exploring Aggregate Asset Price Fluctuations Across Countries: Measurement. C. [17] Brueckner. mimeo. Volume 1B.” Regional Science and Urban Economics. [27] Constantinides. 1559-1695.” Journal of Urban Economics. “Aggregate Investment.. asset prices and economic activity. 107(2). Y. N. E.” Journal of Political Economy. 813-62.J. and C. 125-137.. Leung (2003). North Holland. “Agency costs. [16] Brito. 68.

[39] Ito. 14(1). Moore (1997). Peng (2003). R. Sharpe. (1973). 211-248. West (2004).” Journal of Monetary Economics. Long Swings in Urban Development. and G. E. “Japan land prices: explaining the boom-bust cycle. [33] Edelstein. college aid. 27. Lee (1999). and J. “Japan land prices: towards a structural modeling approach. [41] Kim. J." American Economic Review. Lacayo and D. [38] IMF (2000). Welke (2001). University of California. “Housing.” Journal of Political Economy.. 143-93.” in Koichi Mera and Bertrand Renaud ed. 87-105. N. 18. land prices and growth. 6..E. Berkeley.[28] Cooley. N. [30] Diamond. (1974). (1976). "National debt in a neoclassical growth model. S. K.” International Economic Journal. 15. 55. Iwaisako (1996).-S. Edelstein. Washington D. (1965). Lee (2001). C. A. Asia’s Financial Crisis and the Role of Real Estate.C. “Real estate income and value cycles: a model of market dynamics. W. R. Urosevic. 29 . [29] Deaton. Y. Laroque (2001). “Bank Lending and Property Prices in Hong Kong. R.” NBER Working Paper 10909. [42] King. 69-95. [36] Hanushek. Frontiers of Business Cycle Research. and T. “Land prices and business fixed investment in Japan.J. and J. and K. M. [40] Kandel.-M. Princeton. 1126-1150. Leung and K. Princeton: Princeton University Press.: Princeton University Press.-M. G.: International Monetary Fund.12/2003.” HKIMR Working Paper. Y. mimeo. [43] Kiyotaki. “Asset price and current account dynamics. T. and intergenerational mobility. ed.” Journal of Real Estate Research. New York: M. T. “Explaining Asset Bubbles in Japan. (1995). “Asset returns and intertemporal preferences. “Credit cycles.. [37] Hildenbrand. Stambaugh (1991).” Journal of Economic Growth.” Haas School of Business. S. and W. Core and equilibria of a large economy. and J. A. B. N. 85-108. [35] Gottlieb. E. [32] Edelstein. Peterborough Abbey 1086-1310: A Study in the Land Market. World Economic Outlook. P. 105. Yilmaz (2004). [44] Kiyotaki. Paul. Cambridge: Cambridge University Press. and R. New York: National Bureau of Economic Research. “Borrowing constraints. Paul (2000).” Bank of Japan Monetary and Economic Studies. No. K.” NBER Working Paper 10711. 29-71. [34] Gerlach. [31] Dokko.

“The equity premium: a puzzle. "What drives the property price-trading volume correlation? Evidence from a commercial property market. P. 183-190. 23. [52] Leung. K. Property and Social Transition. 14. J. [49] Leung. K. “Expected Returns and Habit Persistence.” Journal of Economic Literature. [57] Mountford. and E.” The Economic Journal. The Origins of English Individualism: the Family. [55] Macfarlane. Handbook of Population and Family Economics. 567-585. S. N. (2002). New York: Cambridge University Press. E. 12. G. (2004).[45] Kocherlakota. and A. “The Impact of Speculation on House Prices in the United Kingdom. [48] Lau. 1701-1727. Land Prices. A. 411-439. [53] Levin. R. J.” mimeo. (1996).” Economic Modelling.. Y. Feng (2004). K. and Collateral Effects on Economic Fluctuations: Evidence from Japan. 6. 70. 175-203. 13.” Journal of Economic Growth. “Economic growth and increasing housing price. “A stochastic model of mortality. 249-267. C. S.-H. (2003). University of London. Royal Holloway College.” Macroeconomic Dynamics. [46] Kwon. [50] Leung. F. K. and human capital investment. 15. 145-161. 861-99. (1998).” Journal of Monetary Economics.” Journal of The Japanese and International Economies. 42-71. 107. (2002). A. J. Wright (1997). C. (1978). S. C. C. [47] Laitner. and D. [58] Muellbauer. [59] Kalemli-Ozcan. Lau and Y.” Journal of Development Economics. [51] Leung. and R.” in M. [60] Kalemli-Ozcan. “Intergenerational and interhousehold economic links.” Review of Financial Studies. forthcoming. K. Y. Murphy (1997). (1997). Y. Amsterdam: Elsevier. 748-757. [56] Mehra. E.” Journal of Real Estate Finance and Economics. “A Global analysis of an overlapping generations economy with land. [54] Li. (2003). C. 34. “The equity premium: it’s still a puzzle.” Journal of Real Estate Research. Y. “Monetary Policy. Stark ed. 7. (2002). (2001). Y.” Pacific Economic Review. C. Rosenzweig and O.” Journal of Housing Economics. 253-263. 14(3). E. “Macroeconomics and Housing: a review of the literature. “Further inspection of the stochastic growth model by an analytical approach. “Booms and busts in the UK housing market. fertility. 8. “Testing alternative theories of the property price-trading volume correlation. 30 . Leong (2002). “Does the mortality decline promote economic growth. 103-118. Prescott (1985).

” Journal of Real Estate Research.” Quarterly Journal of Economics. “The overlapping generations model of fiat money. [68] Renaud. 79(3): 253-70. M. (1995). and S. (1995). and Booms from Colonial Days to the Present Time. and K. 31 . 50.). 6(2). 1270-92. 110. [71] Schlauch. Handbook of Labor Economics. [69] Sakolski. 1. The Great American Land Bubble: The Amazing Story of Land-Grabbing. Harl Ryder and David Weil (2000). “Internet Cluster Emergence. H. 1-23. [65] Ortalo-Magne. “The 1985 to 1994 global real estate cycle: an overview.. Sebnem.” Papers in Regional Science. and economic growth. Vol. K. 21(1-2): 43-54. B. 44(4-6): 1032-44.” Journal of the Japanese and International Economies. “Housing market fluctuations in a life-cycle economy with credit constraints. K. “Prices and Trading Volume in the Housing Market: A Model with Down-Payment Effects. J. R. [62] Ogawa. New York: Harper & Brothers Publishers. Speculations. D. 13-44. (1932). O. B. Layard.” Graduate School of Business. A. and S.[61] Kalemli-Ozcan. Suzuki (1998). [76] Wheaton.” Journal of the Japanese and International Economies. and S.” Journal of Asian Economics. Suzuki (2000). (1980). K. 12. Research Paper 1501. In: Ashenfelter.” in J. 14. 191-211. W. (2000). “Land value and corporate investment: evidence from Japanese panel data. [73] Singh. 379-406. Land Market: Theory and Practice in Rural India. “Cyberspace and Physical Space in an Urban Economy. search. (2000). J. 232-249. (Eds. [70] Sato. [63] Ogawa. "Labor Supply of Men: a survey". Laposa (2001). [75] Wallace. (1990). Models of Monetary Economies. [72] Shibusawa. “Market valuation and the q theory of investment. [74] Stein. [67] Quah. N.” Japanese Economic Review. 1-21. 5. and prices in a housing market matching model. Stanford University. [64] Ogawa.” Journal of Real Estate Literature. “Vacancy. human capital investment. (1982). “Bubbles in Japan’s Urban Land Market: An Analysis. A. [66] Pencavel.. (1986). Rady (1998). India: Agricole Publishing Academy. Wallace (eds).” Journal of Development Economics. 62.” Journal of Political Economy. (1997). “Mortality decline. Kareken and N. Kitasaka (1999). 153-76. and K. “Demand for bank loans and investment under borrowing constraints: a panel study of Japanese firm data. North Holland. Minneapolis: Federal Reserve Bank of Minneapolis. K. NY. 98.” European Economic Review. F. “E-tailing and Internet-Related Real Estate Cost Savings: A Comparative Analysis of E-tailers and Retailers.

27.” Real Estate Economics. (1999). “Real estate "cycles": some fundamentals. 32 . W. 209-230.[77] Wheaton.

A A. +qt (1 + θ) −1 1−σ qt (qt )2 µ µ Clearly. · −1 + qt 1 − σ qt Thus. we can group terms and further simplify that as "µ ¶−σ/(1−σ) w dqt+1 1/(1−σ) = (1 + θ) −1 dqt qt µ ¶−1/(1−σ) ¶µ ¸ w σ w + −1 .1 Appendix Proof of (13) µ ¶−σ/(1−σ) w −1 . 1−σ qt (qt ) Now notice that µ ¶−σ/(1−σ) µ ¶−1/(1−σ) µ ¶ w w w −1 = −1 −1 . It gives the following expression: ¶−σ/(1−σ) w −1 qt µ µ ¶¸ ¶−σ/(1−σ)−1 ∙ ¶µ −σ d w w 1/(1−σ ) . qt Recall that (12) is qt+1 + r = qt (1 + θ) 1/(1−σ ) Now we differentiate both sides with respect to qt . and dqt q = −w/ (qt )2 . we have dqt+1 = (1 + θ)1/(1−σ) dqt dqt+1 = (1 + θ)1/(1−σ) dqt ¶−σ/(1−σ) w −1 qt µ ¸ ¶µ ¶−1/(1−σ) ∙ −σ −w w 1/(1−σ ) . we can further simplify it as 33 . +qt (1 + θ) −1 1−σ qt dqt qt ³ ´ d w Notice that −σ/(1 − σ ) − 1 = −1/(1 − σ ). Substituting all these t into the expression. qt qt qt µ ¶−1/(1−σ) dqt+1 w 1/(1−σ) = (1 + θ) −1 dqt qt ¶ ¸ ∙µ ¶ µ w w σ .

t+1 − a = Rt + a (1 + θ) c1. σ 6= 1.t )−σ − a(1 + θ)−1 (c2.t µ µ qt w − qt ¶ . at the steady state. it delivers c2.t ) = (c1.t+1 − ac1. −1 −a = qt 1+θ 34 . 1−σ 1−σ where 0 < a < 1.t+1 − ac1.t > 0. the utility function takes the following form u(c1. Now.t )−σ = Rt . −1 q Re-arranging terms.t )−σ which means that ¶σ c2. In this case. and v (c2. Subtractive Habit Formation This form of time-non-separable utility can be at least traced back to the seminal work of Constantinides (1990). we have µ ¶−σ/(1−σ) w 1/(1−σ ) .t+1 qt+1 + r qt+1 + r qt = = • = Rt c1. Thus. we recall (12). dqt+1 = (1 + θ)1/(1−σ) dqt which is indeed (13). c1.which can be shown to be equal to the following expression. we call that value q .2 Proof of (14) Again.t ) = . if q > 0 =⇒ q + r > r.t )1−σ . µ µ ¶ ¸ ¶−1/(1−σ) ∙ w w −1 (1 − σ ) −1 .3 Proofs of the Cases of Non-separable Utility Functions 1.t+1 . qt+1 = qt . q (1 + θ) 1/(1−σ) which means that µ ¶−σ/(1−σ) w > r.t w − qt qt w − qt à Rt µ !σ ¶−1 w Rt + a . It is further imposed that c2.t+1 − ac1. Equation (18) can be re-written as (c1. −1 qt qt A. σ > 0. we will have (14). −1 q + r = q (1 + θ) q Now. and Thus.t+1 − ac1.t )1−σ (c2. A. (1 + θ)−1 (c2. For simplicity.

the left hand side is not well defined for Rt < a (w/qt − 1) .t .t+1 (c1.t+1 . σ 6= 1.t ) = . Overland and Weil (1997.t+1 (c1. Clearly. then c2. θ) (qt ) −σ (w − qt ) Note that the left hand side is equal to ¶σ ¶ µ µ c2. there might and might not be a solution. 2. it is linear in Rt . c1.t )−σ − γ (1 + θ)−1 c2.t )−γ +σRt qt (w − qt ) where f1t = (1 + θ) Rt (qt ) µ ³ ¶σ ´−1 £ ¤−1 −1 w −1 −a ∗ Rt (qt )−1 − a(w − qt ) = (1 + f2t = (1 + θ) (qt ) −σ Rt qt − 1 µ ³ ¶σ−1 ´−1 −1 w −1 −1 Rt qt − 1 −a . Clearly. and the reference therein. whether the steady state exists.29 it is easy to show that dqt+1 f1t = .Notice that with any fixed level of qt . It is clear that if c1. u(c1. 0 < σ . if Rt > 1. ¡ ¢−σ (1 + θ)−1 c2.t+1 c2.t+1 γ (1−σ) . we have dqt+1 /dqt > 1 (which means monotone and explosive). there is a unique value of Rt which solves (19).t ) −γ (1 + θ) c1. c2. For Rt > a (w/qt − 1) > 0. (c1.30 In this case. except when σ is an integer. Equation (18) can be re-written as ¡ ¢−σ (c1. Carroll (2000). given a value of qt and for σ > 1. but when f2t < 0.t )1−σ . dqt f2t −1 −1 −2 (29) µ ³ ¶σ−1 ´−1 w Rt qt − 1 −a > 0. The right hand side is positive even when Rt = 0. see Carroll.t+1 (c1.t+1 (c1. both sides of (19) are increasing in Rt .t )−γ > 0.t ) = 1−σ 1−σ where 0 < γ < 1. then f1t > f2t .t )−γ (c1.t+1 > 0. when f2t > 0. and v (c2. depending on the combination of the parameter values. In fact. and Furthermore. Thus. the left hand side is concave/linear/convex in Rt if σ < 1/ σ = 1/ σ > 1.t )−γ −1 = Rt . For σ < 1. with a positive slope (1 + θ)−1 < 1.t 29 30 For any given set of parameter values. For instance.t )−γ c2.t )−γ (c1. or whether it displays any oscillatory or monotonic dynamics can only be verified numerically for plausible parameter values. the existence can be easily verified numerically. Multiplicative Habit Formation This form of habit formation has also be widely used in the asset pricing and macroeconomics literature. Assuming that there is a solution for (19). 35 .t c1. the utility function takes the following form ¢1−σ ¡ c2. On the other hand. 2000).t+1 (c1. dqt+1 /dqt is negative and |dqt+1 /dqt | can be either larger than unity (which means oscillatory and explosive) or smaller than unity (which means oscillatory and stable).

t+1 . The first order conditions (4) and (5) are still valid.c1. we also consider both subtractive and multiplicative cases.t+1 qt+1 + r qt+1 + r qt = = • = Rt c1.t be the Langrange multipliers of the constraints (1) and (2) respectively.t is the average consumption level of the period t.t+1 ≥ 0.t as given. max u(c1.t+1 .t+1 1 1+θ (30) which is different from both the time-separable case (see (9)) and habit formation case (see (18)). while the right hand side is equal to zero when Rt = 0. c2.t w − qt qt w − qt µ thus we can rewrite the earlier equation as " (1 + θ) (qt ) (w − qt ) σ γ (1−σ)−σ µ ¶−1 w −1 . Condition (3) is also valid because the agent takes the average consumption c1.t . 3. whether the steady state exists.t = c1. ¡ u0 (c1. ¢ ∂v(c2.t ) = qt ∂c2. let λ1.t ). To be comparable with the subtractive habit formation case. To fix the idea.t . it means that the agent maximizes the life-time utility µ ¶ 1 v (c2. delivers a modified version of (6). ³ ´−1 + Rt qt dqt qt w − qt w 1 + γ qt − 1 |{z} +ve | {z } +ve where Rt qt = qt+1 + r by definition. As before.and c2. c1. λ2. Notice also that the left hand side is always positive. dqt+1 /dqt can be either positive or negative. Notice that the left hand side of (20) is invariant to Rt and the right hand side is monotonic increasing (decreasing) in Rt if σ < 1 (σ > 1). Hence. Thus. the utility functions are 36 . and the fact that c1. The consumption in both periods are required to be non-negative. Taking log on both sides of (20) and utilitize the fact that dRt /dqt = (dqt+1 /dqt − Rt ) /qt . combining the first order conditions and the market equilibrium conditions. or whether it displays any oscillatory or monotonic dynamics can only be verified numerically for plausible parameter values. qt where Rt = (qt+1 + r)/qt . Therefore.t ) + 1+θ subject to the constraints (1) and (2). it is easy to show that ⎡ ⎤ µ ¶ 1 − σ dqt+1 1 γ (1 − σ ) − σ ⎥ ⎢ γw(w − qt )−2 = −⎣ ⎦.t ) qt+1 + r ≡ Rt . where c1. Subtractive Catching Up with the Joneses Formally. qt 1+γ ¶−1 #−1 w −1 = (Rt )1−σ .t . c1.

or whether it displays any oscillatory or monotonic dynamics can only be verified numerically for plausible parameter values. the existence can be easily verified numerically. the utility functions 1−σ 1−σ (c2. there might and might not be a solution. ¡ ¢−σ (1 + θ)−1 c2. For ³ ´ w − 1 . It is further imposed that c2. 2.t+1 > 0.t+1 (1 + θ) c1. Multiplicative Catching Up with the Joneses To be comparable with the subtractive habit formation case. with qt being fixed and for σ > 1. We assume that there is a solution for (22). For σ ≤ 1. with a positive slope (1 + θ)−1 < 1.t+1 (c1.t .t+1 .t −σ 1−σ −γ 0 < γ < 1.ac1. Equation (30) can be re-written as µ ¶σ c2.t )−γ c2. Again.t > 0.t )γ (1−σ) . µ ¶−1 w −1 . 0 < σ . 0 < a < 1. 37 . c1. the left hand side is not well defined for Rt < a qt − 1 . c ) = . In except fact. c2.t )−γ ) . and v (c2.31 As in the subtractive habit formation case.t )−σ = Rt .t+1 (c1. σ 6= 1.t w − qt qt w − qt 31 Since the left hand side is equal to µ ¶σ (c1. On the other³hand.t )−γ (c1.t ) = (c11 . it is linear in Rt . and v ( c . it can be further simplified as à µ !σ ¶−1 w Rt Rt .t ) .t+1 − . It is clear that if c1.t ) = (c2. c2. c1. with any fixed level of qt . depending on the combination of the parameter values.t 1−σ 1−σ And similar to the case of subtractive habit formation.t+1 −ac1. whether the steady state exists.t+1 (1 + θ) − a = Rt . (31) −1 −a = qt 1+θ which is different from (19) by only a constant term.t and c2. Equation (30) can be re-written as (c1. there is a unique value of Rt which solves (22). Clearly.t+1 (c1. qt For any given set of parameter values. ´ w when σ is an integer.t ) > 0.t ) assumed are again similar: u(c1.t ) assumed to be similar: u(c1. 4. both sides of (22) are increasing in Rt . The right hand side is zero when Rt = 0.t ) = (c11 .t+1 1. −σ 1−σ 0 < σ. the left hand side is concave/linear/convex in Rt if σ < 1/ σ = 1/ Rt > a q t σ > 1.t+1 qt+1 + r qt+1 + r qt = = • = Rt c1.

and combining this with (32) delivers (28) qt+1 ∙ ¸¾1/(σ−1) ½ β0 1 σ β = (c1.t only. 38 . it is easy to show that fact that dR dqt dqt µ 1−σ Rt qt ¶ dqt+1 1 γ (1 − σ ) − σ = − . c1. and (28) describes a very non-linear relationship between qt and qt+1 . whether the steady state exists. dqt qt w − qt |{z} | {z } +ve +ve t+1 can be either positive or negawhere Rt qt = qt+1 + r by definition.the earlier equation can be re-written as (1 + θ) (qt )σ (w − qt )γ (1−σ)−σ = (Rt )1−σ . or whether it displays any oscillatory or monotonic dynamics can only be verified numerically for plausible parameter values. we have β (c1t ) = qt+1 + r . Notice that by (26).t β ) − − r. Hence. we have We can combine (24) with (27). = (c2. β 0 are functions of c1. The left hand side of (23) is invariant to Rt and the right hand side is monotonic increasing (decreasing) in Rt if σ < 1 (σ > 1).t )σ qt β 1−σ (with the argument of β suppressed). A. and β . qt β 1−σ Notice that under this regime.t+1 )σ β β0 1 . dq dqt tive.t = w − qt . c2. Notice also that the left hand side is always positive. Thus. Thus.t+1 (32) We substitute this into (25) and then we have ¸ ∙ (c2. Taking log on both sides of (23) and utilitize the t+1 t = q1t dq − Rt .4 Some Derivations for the Case of Endogenous Survival Probability β 0 (c1t ) = a1 a0 exp (−a1 c1t ) = a1 (a0 − β (c1t )) . while the right hand side is equal to zero ³ when Rt´= 0.t+1 ) − β (c1. which is only different from (20) by one term. the right hand side of (28) is in terms of qt only.

Figure 1 Dynamics of Land Prices (1a) unstable ( 0 < σ < 1 ) (1b) oscillatory convergence ( σ > 1 ) (1c) center ( σ > 1 ) (1d) oscillatory divergence ( σ > 1 ) Figure 2 Numerical Examples for “oscillatory convergence” and “cycles” (2a) oscillatory convergence (2b) cycles .

gamma=0.36) 100 90 80 70 60 sigma 50 40 30 20 10 0 0 10 20 30 40 50 theta 60 70 80 90 100 Unstable Oscil cvg .04) 100 90 80 70 60 sigma 50 40 30 20 10 0 0 10 20 30 40 50 theta 60 70 80 90 100 Unstable Oscil cvg Figure 3b Habit Multiplicative (r=0.04.3333) 100 90 80 70 60 sigma 50 40 30 20 10 0 0 10 20 30 40 50 theta 60 70 80 90 100 Unstable Oscil cvg Figure 3c Habit Subtractive (r=0.04. w=0.Figure 3a Benchmark (r=0.

Figure 3d Jones Subtractive (r=0. w=0.3333) 100 90 80 70 60 sigma 50 40 30 20 10 0 0 10 20 30 40 50 theta 60 70 80 90 100 Unstable Oscil cvg . gamma=0.04.36) 100 Unstable Oscil cvg 80 60 sigma 40 20 0 0 20 40 theta 60 80 100 Figure 3e Jones Multiplicative (r=0.04.

3.0001) . w=0.Figure 4 Dynamics of Land Price with Endogenous Survival Probability (a) Stable cycle (r=0.5. tau1(t=1)=0. tau1(t=1)=0.5.04.02.0001) (b) Multiple steady states (r=0. sigma=2. w=0. sigma=8.