Residential construction and population growth in New Zealand: Andrew Coleman and Özer Karagedikli. January 2018

Transcription

1 DP208/02 Residenial consrucion and populaion growh in New Zealand: Andrew Coleman and Özer Karagedikli January 208 JEL classificaion: R2, R3 Discussion Paper Series ISSN

2 2 DP208/02 Residenial consrucion and populaion growh in New Zealand: Andrew Coleman and Özer Karagedikli Absrac Beween 996 and 206 Auckland s populaion increased by 499,000, or by slighly more han he increase in he res of New Zealand. Ye only half he number of building permis were issued in Auckland as in he res of he counry. To undersand his difference, his paper uses regional daa o invesigae how populaion growh affecs residenial consrucion. I esimaes ha if Auckland had buil houses a he same rae as he res of he counry (adjused for populaion growh) i would have needed o have buil an addiional 40 55,000 dwellings during he period and needed nearly 9000 more consrucion secor workers. The shorfall was modes unil 2005, bu sharply acceleraed due o he cessaion of aparmen building in cenral Auckland. The resuls show he large increase in he average size of dwellings was no a major facor in Auckland s shorfall relaive o he res of he counry as new dwellings were smaller in Auckland han elsewhere. The views expressed in his paper are hose of he auhor(s) and do no necessarily reflec he views of he Reserve Bank of New Zealand, and he Produciviy Commission. Andrew Coleman: Produciviy Commission and Oago Universiy, and Özer Karagedikli: Reserve Bank of New Zealand. We hank, wihou implicaion, Paul Conway, Roderick Deane, Gary Hawke, Enrique Marinez-Garcia, Anella Munro, Chris Parker, Peer Redward, Tom Smih and he seminar paricipans a he Reserve Bank of New Zealand, Produciviy Commission, Universiy of Oago and Auckland Council for valuable commens. Reserve Bank of New Zealand, 2 The Terrace, Wellingon 60, Po Box 2498, Wellingon 640, New Zealand. Phone (64) ; Fax (64) ; ozer.karagedikli@rbnz.gov.nz. ISSN Reserve Bank of New Zealand

3 The esimaes furher sugges populaion change may be hyperexpansionary as he residenial consrucion demand associaed wih an addiional person is higher han he oupu hey produce. In hese circumsances, populaion increases raise he demand for labour and creae pressure for addiional inward migraion, poenially explaining why migraion-fueled boom-bus cycles may occur. 3

4 4 Non-echnical summary This paper aims o undersand how populaion growh has affeced building aciviy in New Zealand regions during he las weny years. Using panel daa regression echniques, we esimae ha addiional houses are buil for every addiional person in a region. The addiional building permis per person equae o abou 40 m 2 of new consrucion, wih a value of jus over $60,000 in 206 erms. This consrucion is in addiion o he background consrucion ha occurs o replace old houses, which amouns o dwellings per,000 people per year, or approximaely,000 3,000 dwellings per year. The esimaes sugges Auckland s consrucion shorfall beween 996 and 206 was beween 40,000 and 55,000 dwellings, or approximaely 0 percen of Auckland s housing sock. The esimaes of he shorfall are fairly robus o changes in he specificaion of he models; moreover, hey all sugges ha he shorfall was modes unil he end of 2005, afer which i increased rapidly. We also examine he relaionship beween he size of newly consruced dwellings and populaion change. Since four of he sixeen New Zealand regions experienced almos no populaion growh over he period, i is possible o conras he size of newly consruced houses in regions experiencing populaion change wih hose ha did no. These esimaes sugges ha, a leas unil 2005, smaller houses were consruced in growing regions wih above-average incomes, paricularly Auckland and Wellingon, han in growing regions wih below average incomes or in regions wih no populaion growh. This difference appears o reflec he much younger age profile of he residens of Auckland and Wellingon. I appears ha Auckland s housing shorfall was less severe prior o 2005 precisely because of he large number of small aparmens ha were consruced in he ciy. No unil aparmen consrucion almos compleely ceased in 2008 did Auckland s housing shorage sared o become acue. Finally, we analyse he relaionship beween populaion growh raes and he number of residenial consrucion workers. Our esimaes sugges ha a percen increase in populaion growh raes is associaed wih a percenage poin increase in he fracion of he workforce in he consrucion secor. Since regions wih zero populaion growh have percen of heir workers involved in residenial consrucion, each percenage increase in he populaion growh rae increases he number of residenial consrucion workers by approximaely 0 percen. This does no include

5 addiional workers in relaed indusries such as building maerials. Auckland is again an oulier. For mos of he period Auckland had approximaely 9000 fewer consrucion workers han could be expeced from rends around he res of he counry. Clearly, if his shorfall coninues i will be difficul for Auckland o overcome is housing shorage. 5

6 6 Inroducion Rapid populaion growh has been one of he mos sriking feaures of New Zealand s economy in recen years. The migraion-fueled populaion increase, in excess of percen per year, creaed buoyan condiions for New Zealand s consrucion and real esae markes. Real esae prices increased by more han 200 percen in real erms beween 992 and 206, and he fracion of New Zealand s workforce in he consrucion secor increased from 4.8 percen in 992, a pos-970 low, o 7.7 percen in 2009 (prior o he Chrischurch earhquake) and 8.2 percen in 206. However, he populaion increase was no he only facor behind he acive residenial consrucion secor. Beween 99 and 206 he average size of newly consruced houses increased from 32m 2 o 9m 2, a faser rae of increase han in eiher Ausralia or he Unied Saes, he only wo counries for which comparable daa are available. 2 Builders were busy no only because more houses were needed for a larger populaion, bu because people also waned larger houses. Auckland s populaion increase was paricularly large. The number of residens increased by 45 percen beween 996 and 206, or by more han wice as much as he 8 percen increase recorded in he res of he counry (see Table ). Despie an increase in building aciviy, several indicaors sugges insufficien houses were buil in Auckland o keep up wih he populaion increase. Even hough he populaion increased more in Auckland han he res of New Zealand pu ogeher, only half as many new dwellings permis were issued in Auckland as he res of he counry, 53,000 versus 304,000. Moreover, prices increased much faser in Auckland han he res of he counry, by 406 percen in real erms. Finally, he size of Auckland s consrucion secor is smaller as a fracion of he workforce han he res of he counry. To beer undersand how he consrucion secor responds o demand pressures, his paper esimaes how populaion growh has affeced building aciviy in New Zealand during he las weny years. Figure, which plos he relaionship beween populaion growh raes and new residenial building permis per person across sixeen regions, shows he basic idea. 2 New Zealand daa are for 990 and 204; he increase is 43%. US daa are available for he same period and show a 3% increase. Ausralian daa are only available up o 200. See foonoe 3 for sources.

7 Fifeen of he poins lie close o a line wih a slope approximaely equal o one hird, or one new dwelling for every hree exra people. The las poin, represening Auckland, lies far below his line. A simple exrapolaion suggess ha if Auckland had buil a he same rae as he res of he counry, an addiional 50,000 o 60,000 houses would have been buil. The firs purpose of his paper is o refine his esimae using more complex saisical echniques o accoun for a number of economeric issues ha necessiae he use of higher daa frequency. Populaion esimaes for he sixeen regions are available on an annual basis from 996 onwards, bu regressions using annual daa are no paricularly useful because of delays beween when populaion changes and building aciviy occur. These delays can be exacerbaed by capaciy consrains in he consrucion indusry, for when hese bind a large populaion increase can cause a backlog of consrucion aciviy ha akes several years o clear. These consrains sugges he relaionship beween populaion change and consrucion aciviy can be highly variable over shor periods of ime, even if i is sable over longer periods. These delays can also be expeced o induce complex paerns of serial correlaion ino he daa. Using panel daa regression echniques, we esimae ha addiional houses are buil for every addiional person in a region. We conrol for differen combinaions of regional and ime fixed effecs o allow for differen regional paerns and for differen shocks such as he global financial crisis, wih differen lag srucures, and wih differen ways of esimaing sandard errors o explore he robusness of he resuls. The addiional building permis per person equae o abou 40 m 2 of new consrucion, wih a value of jus over $60,000 in 206 erms. This consrucion is in addiion o he background consrucion ha occurs o replace old houses, which amouns o dwellings per,000 people per year, or approximaely,000 3,000 dwellings per year. The addiional consrucion associaed wih a new person is more han New Zealand s per capia Gross Domesic Produc, which in 206 was $54,78. 3 This suggess ha ne inward migraion is likely o be hyper-expansionary, as he immediae demand for housing by immigrans exceeds heir producive poenial. The esimaes also sugges Auckland s consrucion 7 3 Saisics New Zealand. Regional Gross Domesic Produc: Year ended March 206, published March 30, 207.

8 shorfall beween 996 and 206 was beween 40,000 and 55,0000 dwellings, or approximaely 0 percen of Auckland s housing sock. The esimaes of he shorfall are fairly robus o changes in he specificaion of he models; moreover, hey all sugges ha he shorfall was modes unil he end of 2005, when i increased rapidly. The second purpose of he paper is o examine he relaionship beween he size of newly consruced dwellings and populaion change. Since four of he regions experienced almos no populaion growh over he period, i is possible o conras he size of newly consruced houses in regions experiencing populaion change wih hose ha did no. These esimaes sugges ha, a leas unil 2005, smaller houses were consruced in growing regions wih above-average incomes, paricularly Auckland and Wellingon, han in growing regions wih below average incomes or in regions wih no populaion growh. This difference appears o reflec he much younger age profile of he residens of Auckland and Wellingon. I appears ha Auckland s housing shorfall was less severe prior o 2005 precisely because of he large number of small aparmens ha were consruced in he ciy. No unil aparmen consrucion almos compleely ceased in 2008 did Auckland s housing shorage sar o become acue. The hird aim of he paper is o analyse he relaionship beween populaion growh raes and he number of residenial consrucion workers. Our esimaes sugges ha a percen increase in populaion growh raes is associaed wih a percenage poin increase in he fracion of he workforce in he consrucion secor. Since regions wih zero populaion growh have percen of heir workers involved in residenial consrucion, each percenage increase in he populaion growh rae increases he number of residenial consrucion workers by approximaely 0 percen. This does no include addiional workers in relaed indusries such as building maerials. Auckland is again an oulier. For mos of he period Auckland had approximaely 9000 fewer consrucion workers han could be expeced from rends around he res of he counry. Clearly, if his shorfall coninues i will be difficul for Auckland o overcome is housing shorage. 2 A Simple Model of Populaion Growh, Housing Sock and Building Aciviy This secion develops a simple model o explore how building aciviy is affeced by populaion growh and depreciaion. 8

9 Le H = sock of houses a he beginning of period P = populaion a he beginning of period BP = building permis issued beween and + BP = building pu in place a ime beween and + 9 H = he number of houses ha depreciae beween and + α = demand for housing per person, approximaely one hird. The empirical analysis conduced in his paper is based on building permi daa, no he acual amoun of consrucion aking place. In pracice he difference beween hese wo series is small for, as Saisics New Zealand (207) observes, approximaely 95 percen of building permis resul in consrucion, and mos of his consrucion akes place wihin a year of he issue of he permi. 4 Noneheless, he following derivaion makes allowance for his poenial difference by assuming he acual amoun of building aciviy may differ from he amoun of building permis by a random amoun: BP wbp e () The coefficien w is he average fracion of building permis ha are subsequenly consruced. The sock of housing, which is no observed by he economerician, is H (2) ( ) H BP Populaion change is a random variable: n P P P (3) 4 Beween March 998 and March ,49 permis were issued in Auckland and 295,939 in he res of he counry. The number of compleions were 38,229 (93.9 percen) and 280,249 (94.4 percen) respecively. Source: Subnaional dwelling permis and compleion esimaes by Saisics NZ, July 207.

10 0 Le U be a measure of unme housing demand a he beginning of he period. We assume ha he demand for housing per person depends only on he populaion: H P U (4) Suppose α is independen of prices. New housing demand z during a period is a funcion of he populaion change n and he depreciaion of he housing sock ha occurs in he period: H n z (5) Poenial demand during period is herefore ) ( H n U z U U (6) Noe ha ) ) (( ) ( BP U BP z U BP H n P H P U (7) Assume he number of building permis is a linear funcion of poenial demand plus a random disurbance erm: 0 0 v U BP (8) Then v U e v U w BP 0 0 ) ( (9) and

11 ) ( ) ( ) ( ) ( ) ( ) ( ) ( v v U H n v v U z v z v U U v z BP U v z U BP (0) When 2 0, he level of poenial demand will be a saionary variable if he populaion demand shocks n and he shocks o building supply v are saionary, even if building aciviy does no respond fully and insananeously o demand shocks: ) ( ) ( ) ( ) ( ) ) (( ) ( v U z v U U z v U H n H P z BP H n P z U z U () If he building indusry responds immediaely and fully o demand, λ = and he equaion simplifies o 5 v H n BP (2) In his case, a regression of building permis agains he populaion growh rae should recover an unbiased esimae of α, and he residuals of he equaion should be uncorrelaed hrough ime if he idiosyncraic shocks v are uncorrelaed. If λ, he expeced value of he populaion growh regression coefficien will be λα, no α, and he error erm will have wo serially correlaed componens. The firs componen occurs because he unobserved variable U - is no included in he regression. I will have posiive serial correlaion if populaion shocks have posiive serial correlaion or if λ< l and unme 5 If he building indusry responds fully o poenial demand, λ= bu λ0 =/w > o allow for he fracion of issued permis ha are no buil.

12 demand is carried over from period o period (equaion ). The second componen reflecs he negaive moving average componen associaed wih he building supply shocks v. I is negaively correlaed because shocks occurring one period are made up in subsequen periods. The overall serial correlaion of he error process could be posiive or negaive depending on he relaive size of he shocks. As he amoun of unme demand is no observed, he relaionship beween building aciviy and populaion change canno be esimaed in an enirely saisfacory manner. One approach is o regress building permis agains conemporaneous populaion growh, wih lagged populaion growh included as a proxy for unobserved unme demand. The oal effec of populaion growh on building aciviy is found by summing he coefficiens on differen lags. This sum will be biased downwards if insufficien lags are used. An alernaive approach is o aggregae he daa ino fewer bu longer periods, say wo-year periods or four-year periods insead of one-year periods. We do boh, alhough prefer aggregaing he daa ino longer lengh horizons. As he observaion period is lenghened, more of he building secor s response o demand shocks occurs wihin he conemporaneous period and he coefficien beween building permis and populaion change will be closer o he rue long run occupancy raio α. Noneheless, he coefficien will sill be downwardly biased as some of he building associaed wih he populaion increase aking place a he end of he period will ake place in he subsequen period. The bias can be calculaed for differen values of he parameer λ. Consider aggregaing wo periods, and +, ogeher. I follows from equaions 0 and (recalling ha z n H ) 2 BP z ( ) U ( v v ) BP z z z ( ) U ( ) z ( ) z ( v v ) ( ) U 2 ( ) U v v ( v v v ) ( ) v (3) Adding he wo periods ogeher,

13 BP BP ( )(2 ) U z v ( ) v 3 ( ) z (2 ) v (4) If a hird period were added BP BP ( ) BP z v ( 2 ( ) z ( ) ( ) 2 ( ( ) ( ) U 2 ) v ( ) v ( ( ) ( z ) 2 v (5) The formula can be readily exended o longer periods. Equaion 5 can be inerpreed as follows. When hree years are combined ino a single period, he fracion of he desired building aciviy aking place wihin he hree years depends on he years in which he populaion increase occurs. If he populaion increase akes place in he hird year, only a fracion λ of he new houses will be buil by he end of he combined period. If he populaion increase akes place in he second year, a fracion λ(+(-λ)) of he new houses will be buil by he end of he combined period. If he populaion increase akes place in he firs year, a fracion λ(+(-λ) +(-λ) 2 ) of he new houses will be buil by he end of he combined period. The average response can be calculaed as an average of he differen response over he hree years. As he lengh of he combined period increases, i can be shown ha almos all new houses will be buil in he same period as he populaion increase. 6 For his reason, he esimaed coefficien beween he 6 If N periods are joined ogeher, he quaniy of building aciviy associaed wih a se of populaion shocks z occurring from sub-periods 0 o N- will be N S ( z) z ( ( )) z... ( ( )...( ) ) z N N N 2 S N () N I is sraighforward o show lim. I follows ha if z is a saionary sequence, N he average fracion of building aciviy associaed wih a change in populaion ha occurs

14 number of building permis and he populaion change should increase as he lengh of he combined period increases, and approach he rue parameer α asympoically. The raios of he coefficiens esimaed using differen period lenghs depend on he parameer λ. In secion 4 of he paper we aggregae daa ino periods ha vary beween one year and weny years and show he esimaed coefficiens increase as he period lengh increases and appear o converge. By comparing he coefficiens esimaed using differen lengh periods wih he heoreical relaionship implied by equaion 5 (or is equivalen for longer periods), i is possible o ge a sense of he size of he parameer λ. The esimaed numbers sugges ha λ lies beween 0.6 and 0.8, which means percen of he housing demand induced by populaion change is sared wihin a year. If λ = 0.6, and we examine 4-year periods, 84 percen of new houses should be buil in he same period ha he populaion change occurs, wih he res occurring aferwards. If λ = 0.7, he figure rises o 89 percen. These numbers sugges ha when we aggregae he daa ino 4-year periods we are likely o underesimae he number of building permis associaed wih populaion change, by 0 7%. In urn, his means we are likely o underesimae Auckland s building shorfall. Why migh we prefer o choose longer periods raher han esimae a regression using one-year periods and several lags? One reason is pracical: i is difficul o esimae he coefficiens of a large number of lagged variables accuraely if hey are serially correlaed, even wih a lo of daa. A second issue concerns he effecs of capaciy consrains on building aciviy. Whenever an indusry has capaciy consrains ha occasionally bind, he iming of he supply response o differen sized demand shocks will vary. I may ake hree years o respond o an unusually large populaion influx, for insance, whereas he demand from a small populaion increase migh be me wihin he year. If his is he case, he lag srucure beween populaion changes and building permis will no be consan. Moreover, regression esimaes will be biased as he residuals of he esimaed equaions will be correlaed wih he lags of he populaion variable. These problems are reduced by increasing he size of he period, for hen mos of he building aciviy associaed wih differen sized shocks akes place wihin he period. 4 wihin a period approaches.

15 5 3 Daa Sources Saisics New Zealand collecs daa on he size, ype, number and dollar value of building permis. Some of hese daa are freely available from Saisics New Zealand s Infoshare daabase, while ohers have o be purchased separaely. The basic regressions in secion 4 and 5 are esimaed using annual Infoshare daa for he period July 996 June 206 (Daa series: Infoshare BLD3AA). These daa include he number, area and nominal value of building permis issued for new unis, and he number and value of building permis issued for aleraions. The daa cover sixeen differen regions. Daa from 99 o 996 were obained from he same source, bu are no used in mos of he analysis as hey could no be mached o populaion daa. Annual regional populaion figures are available from Saisics New Zealand from 996 (Daa series: Infoshare DPE05AA.) We use esimaes for he populaion a he end of June for each year. In addiion, we use populaion figures from he 99 census o calculae five-year populaion growh raes from 99 o 996. The mos deailed sources of regional age-specific demographic informaion are he Saisics New Zealand censuses, which are available for 996, 200, 2006 and 203. Alhough hese daa provide esimaes of he populaion broken down ino five-year age groups, he irregular frequency of he censuses limi heir usefulness. Noneheless, we combine he census daa wih he annual esimaes of he populaion of four age groups (0 4; 5 39; 40 64; and 65 plus) ha Saisics New Zealand produced for he period o creae age-specific demographic variables. (Daa series: Saisics New Zealand, Subnaional populaion esimaes (RC, AU), by age and sex, a 30 June 996, 200, (207 boundaries).) Saisics New Zealand provided us wih a special compilaion of daa ha gives he number of dwellings disaggregaed by ype (sand-alone houses, aparmens, ownhouses, independen living unis in reiremen villages) in eigh size caegories (< 00 m2; m2; m2; m2; m2; m2; m2 > 400 m2). The daa were compiled for 6 New Zealand regions and 2 wards of he Auckland region, and are available on an annual basis from The daa also include he value of permis, and he oal area of permis. We deflae he nominal value of building permis by he residenial building

16 componen of he Capial Goods Price Index (Daa series: CEP007AA). We do no have separae indices for differen regions. Beween June 996 and June 206 his index increased by 95 percen. The nominal value of aleraions was also deflaed by his series. In secion 6 we examine rends in regional consrucion secor employmen. We exclude workers engaged in heavy engineering consrucion projecs such as roads or commercial buildings by only including he number of firms and employees in he consrucion secors E30 and E32. 7 (Daa series: Saisics New Zealand Business Demographic Saisics Geographic unis by region and indusry ) 4 Esimaing he Effec of Populaion Growh on Building Permi Numbers In his secion we analyse he relaionship beween populaion growh raes and he number of building permis. We firs esimae equaions ha vary in erms of he lengh of a period, bu which do no include lagged populaion growh raes. In secion 4.2 we esimae models ha include lags, and in secion 4.3 we esimae models ha incorporae addiional demographic informaion Models excluding lagged populaion growh We esimae he following equaion: BP h i region dummy h 0 h 4 h Popn i e h i i ime dummy h 2 Chrischurch h 3 (6) where h BP i is he number of building permis per capia in region i during period, and h refers o he lengh of he period in years; Chrischurch is a dummy variable equal o for ime periods afer he 200 earhquake, and zero oherwise; and 7 Daa are only available for 5 regions as Tasman and Nelson are combined.

17 7 Popn i is he populaion change in region i during period, as a fracion of he iniial populaion. In each case he regression included a full se of regional and ime dummy variables. 8 The regressions are esimaed separaely eiher using he 5 regions excluding Auckland or he 6 regions including Auckland. The daa are aggregaed ino differen lengh periods, where he lengh of he periods, h, is equal o eiher, 2, 4, 5, 0 or 20 years. We es he residuals for firs order serial correlaion for period lenghs less han or equal o five years. 9 We do no esimae he serial correlaion srucure if he period lengh is en or weny years as here are insufficien observaions. Table 2 conains he level and sandard errors of he esimaed coefficiens ˆ 4, he R 2 of he regression, and he esimaed serial correlaion of he errors for regressions esimaed using differen period lenghs. The coefficiens are esimaed using ordinary leas squares and he sandard errors are calculaed using he Huber -Whie mehod. The op panel in able 2 repors he resuls when Auckland is excluded from he regressions. The main coefficien of ineres, 4, on he variable Popni is smalles for he shor lengh periods (h =, 2), and ges larger as he lengh of he period increases. When he period lengh is shor, he esimaed errors have saisically significan posiive correlaion. We suspec his is due o capaciy consrains in he building secor: in he shor-erm a large increase in populaion will generae large amouns of consrucion in successive periods as no all of he demand will be me immediaely. The posiive serial correlaion in he error srucure will be exacerbaed if populaion growh raes have posiive serial correlaion. A longer horizons, however, here may be negaive serial correlaion in he error srucure, as builders compensae for under or over supply in earlier periods. For example, if consrucion aciviy is less han required during a five-year 8 The equaions were also esimaed wihou regional dummy variables, in which case he sandard errors were calculaed using Thompson clusered sandard errors. As he regional dummy variables were ypically saisically significan, we kep hem in our preferred specificaion. 9 If he esimaed errors are serially correlaed, we also esimaed he equaion using feasible general leas squares esimaion. The resuls are similar and are available from he auhors upon reques.

18 period i may be made up in he subsequen five years, generaing a negaive serial correlaion paern in he residuals. The residuals of our esimaed regressions have hese paerns. The residuals of he one or wo-year period regressions have large and saisically significan posiive serial correlaion, consisen wih he exisence of capaciy consrains. However, he residuals of he five-year period regressions are negaively correlaed, and significan a he 0 percen confidence level. The residuals of he four-year period regressions only have small and saisically insignifican serial correlaion, possibly because he wo forces offse each oher. The boom panel in able 2 repors he resuls when Auckland daa are included in he regressions. The wo ses of resuls are almos idenical excep for he weny-year period regressions, when here is only a single observaion for each region and fixed regional effecs canno be included. In he shorer period regressions, he inclusion of regional and ime dummy variables mean he coefficien on he populaion growh variable is esimaed from wihin-region variaion in populaion growh raes and building permi numbers, no across regional variaion, and he inclusion or omission of Auckland makes lile difference o he resuls. Including Auckland in he weny-year period regression reduces he esimaed coefficien ˆ 4, as building aciviy in Auckland is much lower han can be expeced from he paern in oher regions (see Figure ). We again focus on he four-year period regressions as he residuals appear serially uncorrelaed. The esimaed coefficiens ˆ 4 in he 4-year horizon regressions are and respecively when Auckland is excluded or included, implying ha new housing permis are issued for each addiional person. The coefficiens are precisely esimaed wih sandard errors of and respecively, and he overall fi in hese regressions are 0.88 and Even if hese coefficiens underesimae he long run effec of populaion growh on building aciviy by a sixh, because some consrucion occurs afer he period ends, less han 0.3 houses are buil for each new person. This suggess ha he marginal effec of populaion change on building aciviy is less han he average housing/populaion raio. This populaion relaed-consrucion is over and above consrucion ha replaces old or depreciaed houses, which is capured by he consans. The consans vary across regions bu he average value is beween 2.5 and 3.0 building permis per 000 populaion. We use he esimaed coefficiens from he able 2 o calculae he 8

19 accumulaed housing shorage in Auckland based on he relaionship beween building aciviy and populaion increases in he res of he counry. The las column in able 2 shows he esimaed accumulaed shorage as of 206. Wih one excepion, he models sugges Auckland s housing shorage is beween 38,000 and 59,000 houses. 0 The esimaed shorage in 206 of our preferred four-year horizon model is 54,000. Figure 2 shows he esimaed accumulaed shorage in Auckland for he period made using he 4-year horizon regression coefficiens. The shorage was modes unil Beween 2004 and 2005 he housing shorage reduced, reflecing a subsanial increase in aparmen building (his is discussed in more deail in secion 5). However, he shorage is esimaed o have increased sharply afer The Auckland housing shorage we esimae is larger han he esimae of 20,000 o 30,000 cied in Spencer (206). Par of he difference reflecs a difference in mehodology: he esimae in his paper calculaes he shorfall under he assumpion ha populaion changes in Auckland have he same effec on he demand for housing as hey do in he res of he counry. This would no be he case if people in Auckland waned higher occupan/housing raios han people in he res of he counry, possibly because Auckland has a younger populaion or because house prices are higher. Moreover, demoliion raes could be lower in Auckland han elsewhere because he average age of he housing sock is lower, which would produce lower esimaes of he shorfall. The accumulaed shorfall is also calculaed relaive o 996, an earlier (and more explici) year han some oher esimaes. Noneheless, he esimaes made in his paper have a lower marginal occupan/dwelling raio han hose ofen used o esimae Auckland s housing shorage, as he esimaes are based on he marginal raher han average occupan/housing raios prevailing in oher regions of he counry. 9 0 The esimaed shorfall using he regression ha uses weny-year period daa excluding Auckland is 84,500. This regression only has 5 observaions, and he coefficien is esimaed using cross-region no wihin-region variaion. Auckland s housing sock is younger on average han he res of he counry because he ciy has grown rapidly, which migh sugges demoliion raes are lower (and ha we overesimae he shorage because fewer houses are knocked down han in he res of he counry.) However, he pressure for inensificaion is higher in Auckland han elsewhere, which migh resul in more demoliions as developers seek o build muliple unis on valuable secions.

20 Models including lagged populaion growh For robusness purposes, we esimae he equaion wih one, wo or hree lags of populaion growh in addiion o he conemporaneous period populaion growh. The coefficiens on even longer lags were small and never significanly differen from zero. These resuls are repored in able 3. The lagged models are esimaed using annual daa as well as daa aggregaed ino wo or four-year periods. As he coefficiens on he lagged populaion growh variables in he models using one and wo-year period daa are posiive and saisically significan, he oal amoun of building aciviy associaed wih populaion growh is larger han esimaed in he one and wo-year horizon period lengh models in able 2. Noneheless, he esimaes sugges mos building aciviy associaed wih populaion increases akes place wihin four years, wih he vas majoriy of ha aking place wihin wo years. In oher respecs, he regressions have very similar feaures o he resuls in able 2. The coefficiens indicae he effec of conemporaneous populaion increases on building aciviy sill increase as he period lengh increases. The firs order serial correlaion paern is almos idenical. The regression using shorer lengh daa have posiively correlaed residuals bu here is no serial correlaion in he residuals esimaed using four-year period daa. In he equaion wih four-year periods, he coefficien on lagged populaion growh is 0.038, bu i is no saisically significan from zero a 5 percen significan levels. The coefficien on conemporaneous populaion growh is The sum of he coefficiens is 0.294, of which 87 percen akes place in he conemporaneous period. This suggess our preferred specificaion in secion 4., which excludes lags, is no badly mis-specified. I may be recalled from he heoreical model in secion 2 ha if 70 percen of housing demand is consruced wihin a year, 87 percen of he housing associaed wih populaion increases ha occur wihin a four year period will be consruced wihin he four years. I hus appears he daa are broadly consisen wih he model in secion 2 wih a parameer λ equal o 0.7. The esimaed Auckland housing shorage is larger a every horizon relaive o he esimaes in able 2. When a lag is included in he four-year period regressions, he esimaed Auckland shorfall in 206 increases from 54,000 dwellings o 65,500. Grimes and Hyland (205) esimaed a model of housing supply and demand

21 using quarerly daa from 72 erriorial local auhoriies over he period Using a panel coinegraion response based around long run supply and demand equaions, hey esimaed a shor-run lagged response funcion based on he indirec response from populaion o prices and from prices o building aciviy. They used he resuls from his model o esimae how long i would ake for building aciviy o respond o an increase in populaion ha ook place over a five-year period. They esimaed ha only half of he building aciviy would occur wihin six years of he sar of he populaion increase, and ha he whole adjusmen would ake nine years. Our esimaes sugges ha building aciviy responds o demand much more quickly han his. Consider he response o 00 people moving ino a region over a fouryear period. Using he four-year period regression, we esimae 26 building permis would be issued wihin he iniial four years, and a furher 4 permis would be issued in he subsequen four-year period, alhough he laer number is no esimaed accuraely and i is no saisically differen from zero. This means over 85 percen of he building permis are issued wihin he iniial four-year period. 2 We suspec he esimaes of Grimes and Hyland subsanially over-esimae how long i akes for he building secor o respond o populaion changes, a leas in areas of New Zealand ouside Auckland Models including age-specific populaion informaion The above models assume ha he number of new dwellings does no depend on he age srucure of he populaion. This assumpion could be imporan, as Auckland s populaion age srucure is quie differen o ha of mos oher regions of he counry. Moreover, here is inernaional cross-counry evidence suggesing ha, a he naional level, he amoun of residenial consrucion depends on he age srucure of he populaion, no jus he oal number of people (Lindh and Malmberg 2008; Monne and Wolf 207). Monne and Wolf, for example, argue ha he key deerminan of residenial consrucion in 20 OECD counries (excluding New Zealand) is he growh in he populaion of people aged Unforunaely, here are significan daa limiaions ha resric us from 2 If 00 people moved ino an area over a year, he one-year period regression indicaes an overwhelming majoriy of he building permis are issued wihin hree years of he populaion increase.

22 comprehensively esimaing how he number of he new dwellings in a region depends on he age-srucure of he populaion. For example, we do no have regional daa on he number of people aged Noneheless, in he appendix we provide resuls ha sugges ha he age-srucure of he new residens in a region did no have a major effec on he oal demand for new dwellings. As we show in secion 5.3, however, hey do affec he size of he dwellings ha are consruced. As discussed in secion 3, only limied regional age-specific daa are available. Prior o 2006, he only daa come from censuses, which resrics us o analysing he relaionship beween building aciviy and populaion change over five-year periods. Four age groups are available: 0 4; 5 39; 40 64; and 65 plus. Table 4 shows he average conribuion of each age group o he populaion change in each region over he weny-year period. Across he counry, 6% of he 960,000 populaion increase occurred in he 0 4 age group, 6% in he 5 39 age-group, 50% in he age-group, and 28% in he over 65 age-group. Auckland s populaion change was very differen han ha which occurred in mos of he res of he counry. Auckland had much larger increases in he number of children and younger aduls (people aged 5 39), and a relaively small increase in he number of people aged over 65. In oal, 46% of Auckland s populaion increase beween 996 and 206 was aged less han 40; in conras, he number of people aged less han 40 declined in 9 ou of he 6 oher regions. Of he oher regions, only Canerbury and o a lesser exen Wellingon had a subsanial increase in he number of people aged under 40. Does Auckland s disincive demographic profile explain why so few houses were consruced in Auckland relaive o is populaion increase? The esimaed regression resuls, which are displayed in able A in he appendix, do no suppor his conenion. Unforunaely, he coefficien esimaes have high sandard errors, due o he high degree of correlaion beween he differen demographic variables. This means in mos circumsances i is no possible o rejec he hypoheses ha individual demographic variables conribue no addiional explanaory power o he regressions, even hough he esimaed coefficiens someimes appear very differen. These regression resuls end o sugges ha he number of new building permis appears o respond o oal populaion change, no he individual componens. The demographic daa are used o see if he populaion age-srucure affecs 22

23 he size of newly consruced dwellings in secion 5.3. These resuls are sronger, and sugges ha regions wih more young people build more small dwellings and more very large dwellings. Our enaive conclusion, herefore, is ha while he age-srucure affecs he size of newly consruced dwellings, i does no have much effec on he overall number of dwellings. In paricular, we have found no evidence ha he young average age of Auckland s new residens can explain why so few new dwellings have been buil. If anyhing, our esimaes sugges ha ignoring age-specific informaion undersaes Auckland s housing shorfall Differences in New Housing Size Across Regions The average size of new dwellings in New Zealand increased from 33 m 2 in 990 o 9 m 2 in 206. This is a faser rae of increase han occurred in eiher Ausralia or he Unied Saes. 3 In Auckland he fracion of newly consruced dwellings smaller han 50 m 2 fell from 68 percen o 32 percen, while he fracion over 250m 2 increased from 8 percen o 26 percen. These figures raise wo quesions: why did he size of new houses increase; and is he increased size of houses a major facor behind Auckland s housing shorage? The answer o he firs quesion is unclear, for he saisical evidence is no srong enough o unangle he relaive imporance of differen facors. In conras, he daa clearly show ha Auckland s housing shorfall relaive o he res of he counry was no primarily because Auckland builders specialized in building large houses. Alhough he size of new housing in Auckland mirrored rends in he res of he counry, more small houses, primarily aparmens, were buil in Auckland prior o New Zealand daa are from Saisics New Zealand, Number, value and floor area by building ype, naure and region BLD075AA. Ausralian daa are from he Ausralian Bureau of Saisics, Building Approvals February (The series is no longer produced.) U.S daa afer 999 are from he websie hp:// spreadshee " SFForSaleMedAvgSqF". Earlier daa are from hp:// This sie has a series of books wih he daa e.g. US Deparmen of Commerce (2000) "Characerisics of New Housing 999".

24 5. The size disribuion of new consrucion: a heoreical overview The classic analyses of he markes for differen qualiy houses were developed separaely by Sweeney (974) and Rosen (974). Their analyses show ha housing markes require he simulaneous consideraion of (i) heerogeneous housing qualiy, (ii) a housing demand funcion ha depends on rens, curren house prices, he expeced rae of change of house prices in he fuure, and oher facors such as he number of people in he local housing marke, heir income, ineres raes and axes, (iii) knowledge of how households form expecaions abou fuure house prices, (iv) a supply funcion for new consrucion ha is inelasic and subjec o capaciy consrains, and (v) a rule ha decides he order in which houses differing in erms of qualiy are buil when he demand o build is unusually high. Sweeney s model calculaes a long-run marke equilibrium ha depends on long-run supply and demand facors for houses ha differ in erms of heir qualiy, and a se of ransiion pahs o his equilibrium. He observed ha he demand for one qualiy of housing depends on he prices for all qualiy ypes, as buyers make price/qualiy comparisons and buy he qualiy ype ha offers hem bes value. In he long run, prices mus reflec producion coss o ensure posiive amouns of each qualiy level are supplied. Consequenly, he quaniy of housing of each qualiy ype depends on he demand for each ype of housing when prices are equal o long-run producion coss. New housing is buil for one of hree reasons: (i) o mainain he qualiy of he old sock of housing, via aleraions of he exising sock and replacemen of houses as hey depreciae; (ii) o improve he qualiy of he housing sock in response o changes in demand facors; and (iii) o increase he number of houses in he face of populaion growh or changes in he demographic composiion of he populaion. 4 Unforunaely, very lile can be said on 24 4 In he absence of populaion growh or facors changing he desired qualiy of he housing sock, he qualiy level of new consrucion ends o be higher han he exising sock. This is because a large fracion of he supply of lower-qualiy housing comes from he depreciaion (or filering ) of beer-qualiy housing hrough ime. New lower-qualiy housing is mos likely o be buil in fas growing ciies, for hey lack a sufficien sock of depreciaed older dwellings o house lower-income households. Even in hese circumsances new housing is likely o be disproporionaely beer qualiy, as large numbers of new higher-income (or higher-wealh) residens compee for he available sock of higher-qualiy houses.

25 how he qualiy profile of newly consruced houses depends on fundamenal economic facors, as his profile is essenially he residual beween he qualiy profile of he desired sock of housing and he qualiy profile of he exising sock of housing. Noneheless, housing markes can be away from heir long-erm equilibrium for long periods if here is a large demand shock. During his ime builders slowly aler he number of houses across he enire qualiy disribuion, closing supply-demand mismaches ha are refleced by prices which deviae from heir long-erm levels. 5 New Zealand experienced a large number of economic and demographic changes ha increased he demand for beer-qualiy housing afer 990. There was a subsanial increase in household incomes afer 994, paricularly amongs beer-paid workers. 6 Real ineres raes declined sharply afer 2000, in line wih inernaional rends, reducing he user cos of durable asses including housing. There was a subsanial decline in inflaion following he Reserve Bank Ac (989), leading o even larger declines in nominal ineres raes and hus a reducion in he effecs of morgage il Suppose here is an increase in incomes ha induces a demand for beer qualiy dwellings across he whole qualiy disribuion. This creaes a mismach beween he qualiy of he desired housing sock and he exising housing sock, and prices increase o mach curren demand wih he available supply. The exen ha prices need o increase depends on he exen ha fuure prices are expeced o reduce. When expecaions are raional, and he supply imbalance is small, a small price increase may be sufficien o equae demand wih he available supply, as expeced fuure price declines will reduce conemporaneous demand. If expecaions are no raional, or he demand imbalance is very large, large price increases may be necessary o reduce demand o mach he available supply. When he oal increase in demand is much greaer han he available building capaciy prices can remain higher han ordinary consrucion coss for some ime, raising profi margins. In response o he increases in prices associaed wih he addiional demand, he mos profiable ypes of houses are buil firs: hese are houses a qualiy levels where he gap beween prices and consrucion coss is larges. Second hand prices can remain above long-erm consrucion coss for lenghy bu ulimaely emporary periods of ime as he qualiy composiion of he housing sock is modified. 6 Minisry of Social Developmen (204) Incomes Repor p3. Median household incomes increased by 45% beween 994 and 204. Much of his was due o higher female paricipaion. Real ordinary ime incomes only increased 24% during his ime. 7 Holding real ineres raes consan, an increase in he inflaion rae raises nominal ineres raes and he amoun of nominal morgage repaymens. The real value of paymens made a he beginning of a sandard able morgage is iled o be much higher han he real value of paymens made a he end of he morgage. As he conracual form of morgages prevens households from borrowing o make hese higher nominal paymens, credi consrained households will find hemselves unable o borrow as much as hey would like. A decline in nominal ineres raes will reduce he effec of hese credi consrains and lead o

26 There were significan changes made o he way ha reiremen income accouns were axed in 989, providing a ax incenive o purchase larger houses as owner-occupied housing became a ax advanaged asse class (Coleman 207). Finally, an increase in he fracion of he populaion aged is likely o have increased he demand for larger houses (Coleman 204). While hese facors should have affeced all regions of he counry, he effec of populaion growh on he desired qualiy of houses should have varied across regions. Builders in regions wih low populaion growh mainly had o aler he qualiy profile of he exising building sock. In conras, builders in regions wih large populaion increases had o cope wih increased demand for larger houses from heir incumben populaions, and he increased demand for houses across he whole qualiy disribuion due o he influx of new people. The exen ha he consrucion indusry concenraed on one qualiy level raher han anoher is an empirical maer ha reflecs he differences beween he cos of new housing and he second-hand price of old housing along he qualiy scale. In urn, he exen ha prices remain above long-erm levels depends on he size of he consrucion secor relaive o he housing shorfall. 5.2 Alernaive measures of Auckland s housing shorfall 26 Did Auckland s housing shorfall occur because builders in Auckland consruced unusually large houses over he period? There are several ways his quesion can be answered, all of which sugges he answer is no. Bu fundamenally, he average size of new houses in Auckland would be larger han hose buil in he res of he counry if Auckland had buil unusually large houses. In fac, he average size of new Auckland houses was 6 m 2 smaller han he average recorded in he res of he counry, 77 m 2 versus 83 m 2. Table 5 shows he resuls of a se of regressions ha use differen merics o measure Auckland s housing shorfall. The firs half of he able includes he oal area and he real value of housing permis, he real value of permis for aleraions, and he oal value of permis, new housing plus aleraions. The second half of he able repors he resuls for he number of permis in hree a demand for beer qualiy housing. See Kearl (979).

27 differen size classificaions: small dwellings less han 50m 2, medium sized dwellings from m 2, and large dwellings in excess of 250m 2. The able repors he resuls of he regression used in secion 3, bu modified for differen dependen variables, along wih he forecas Auckland shorfall over he enire period: 27 BP a i a 4 region dummy a 0 Popn i a e a i i ime dummy a 2 Chrischurch a 3 (7) a where BPi is a paricular measure of building aciviy in region i during period. Each regression includes a full se of regional and ime dummy variables. The regressions are esimaed using he 5 regions excluding Auckland, and use daa aggregaed ino eiher hree-year or four-year periods. The able conains he level and sandard errors of he esimaed coefficiens ˆ 4, he R 2 of he regression, he esimaed serial correlaion of he errors, and he esimaed shorfall (or surplus) in Auckland made under he assumpion ha he value of he regional dummy for Auckland is equal o he average regional dummy variable. The coefficiens are esimaed using ordinary leas squares and he sandard errors are calculaed using he Huber-Whie mehod. The firs row of he able shows he resuls when he dependen variable is he number of building permis per capia per year, using four-year inervals esimaed over he period This is he regression repored in row 3 of able 2, and he coefficien has he inerpreaion ha each addiional person in a region is associaed wih he consrucion of addiional houses. By his meric, Auckland s shorfall was esimaed o be 53,400 dwellings over he period In he second row, he dependen variable is he oal area of new consrucion per capia per year. The regression indicaes ha an addiional person is associaed wih an exra 39 m 2 of consrucion, and ha Auckland had an accumulaed shorfall of 9,950,000 m 2 beween 996 and 206. As he average newly consruced house in New Zealand over he period was 83 m 2, his is equivalen o a shorage of 54,400 dwellings, very similar o he previous esimae. The accumulaed shorfall means ha he oal area of permis issued in Auckland was 27 percen less han required o keep up wih consrucion rends