(1)(2)(3)(4)(5)2012 年 10 月 19 日Issue No: 12/18日本經(jīng)濟分析研究報告辨別短期內(nèi)土地價格可能上漲的地區(qū)Naohiko Baba我們在本文中基于日本各縣詳細數(shù)據(jù)對土地價格的決定因素進行了模型分析，并辨別出了中期內(nèi)地價很有可能上漲的地區(qū)。我們的主要結論如下：我們的時間序列分析和橫向分析均顯示，對日本地價影響最大的人口指標是人口紅利（勞動力人口/受贍養(yǎng)人口）。隨著人口紅利的上升（下降），適齡勞動力人口獲得更多（更少）的資源用于建設生產(chǎn)活動，由此在經(jīng)濟體中形成良性（惡性）循環(huán)，地價隨之上升（下降）。人口紅利對于商業(yè)地價的影響大于其對住宅地價的影響。+81(3)6437-9960 高盛證券株式會社Chiwoong Lee+81(3)6437-9984 高盛證券株式會社Yuriko Tanaka+81(3)6437-9964 高盛證券株式會社綜合考慮人口紅利以及勞動力人口人均名義 GDP 等指標，我們的模型可解釋日本地價變動的 95%左右。此外，勞動人口人均名義 GDP 受到人口紅利的顯著影響，這意味著日本地價的唯一最重要影響因素其實是人口紅利。我們對各縣土地溢價的計算顯示，東京的住宅和商業(yè)用地的溢價都最高。東京商業(yè)用地的溢價尤其高，溢價幅度約為大阪（位居第二）的近三倍。這體現(xiàn)出東京商業(yè)活動高度集中的集群效應。我們基于人口紅利前景篩選出了地價可能會上漲的地區(qū)。東京和大阪地價上漲的可能性位居第一，其次是大東京區(qū)（神奈川縣、埼玉縣和千葉縣)、大大阪區(qū)、九州（不包括佐賀縣和長崎縣）、宮城縣和石川縣。投資者不應視本報告為作出投資決策的唯一因素。 有關分析師的申明和其他重要信息，見信息披露附錄，或參閱 年 10 月 19 日日本經(jīng)濟分析Identifying areas where land prices are likely to rise in near futureJapanese land prices are showing signs of ending a downtrend that dates from the early 1990s,when Japans asset bubble burst. In the benchmark land price report published in late September(prices as of July 1), there were substantial increases in the number of locations where residentialand commercial land prices were either flat or higher year on year in the four largest populationcentersTokyo, Kanagawa, Osaka, and Aichi (see Exhibit 1.)Demographics is the fundamental determinant of land prices. In the report we published beforethe release of benchmark land prices in late September , we focused on the productivepopulation ratio (productive population/total population), using momentum in this ratio (differencebetween productive population growth and total population growth) as an indicator of where landprices are likely to rise in the near future. We found that an increase in nationwide land prices isunlikely, but there is scope for a rebound in Tokyo on a comparatively short horizon.Exhibit 1: Clear halt in land price decline in the four biggest population centersPercentage of locations where land prices are rising or flatResidential land pricesCommercial land prices40353025(%)UnchangedRising35302520(%)UnchangedRising201510501510502011201220112012Note: Share calculated after weighted by nominal GDP of each prefecture.Source: MLIT, Cabinet Office, GS Global ECS Research.We have since been asked many questions about the application of our hypothesis to other partsof the country and about the potential for land prices outside Tokyo to rise. In this report wetherefore seek to identify more robust implications by using data for the past 30 years for Japansall 47 prefectures. We combine information from time series data with information from crosssectional data for the 47 prefectures. In seeking to predict near-term price trends, we analyzeresidential and commercial land separately, whereas in our previous report we used the “allcategory” price.The analytical model we present in this report is based on demographics, which represents themost significant structural change in the Japanese economy. The model has strong explanatorypower for historical land prices, so it should be effective in long-term forecasting. However,forecasting the demographics of individual prefectures is much more difficult than forecastingnational demographics because these forecasts need to incorporate population inflows andoutflows for each age segment. In this report we therefore confine ourselves to medium-term(three-year) forecasts based on current demographic momentum.See the September 5, 2012, Japan Economics Analyst: Demographics implies a rebound in Tokyo land price, albeit a fallnationwide.高盛全球經(jīng)濟、商品和策略研究2232012 年 10 月 19 日日本經(jīng)濟分析In addition to price movement rooted in demographics, there may be a spillover from non-fundamental factors. Land prices are asset prices and, as such, are strongly influenced bysentiment. For example, a rise in closely watched Tokyo land prices could change investor anddeveloper sentiment in other parts of the country, leading to an increase in local prices. We aresaving the results of such analysis for our next report.Preliminary considerations (1): Time series observationsBefore commencing our model analysis, we set out our basic data . Exhibit 2 shows how majorpopulation centers experienced a massive bubble in land prices from the late 1980s to early1990s, albeit to different degrees, and a subsequent, prolonged correction that is now beginningto show signs of ending. The correction has been more severe for commercial land, which rosemore than residential land during the bubble.Exhibit 2: Decline halting for residential and commercial land pricesResidential land priceCommercial land price10009008007006005004003002001000(Thousand yen/sqm)TokyoOsakaKanagawaAichi9000800070006000500040003000200010000(Thousand yen/sqm)TokyoOsakaKanagawaAichi1975198019851990199520002005201019751980198519901995200020052010Source: MLIT.The key underlying factor is a demographic trend we discussed in our previous report. There aremany demographic indicators. Exhibit 3 picks out the so-called demographic dividend. Thedefinition of demographic dividend is productive population divided by dependent population,where the productive population is ages 15 to 64 and the dependent population is all other ages.It is a measure of how many persons in the productive population are shouldering the burden ofproviding services (pensions, healthcare, etc.) for the elderly and raising childreni.e., providingfor two segments of the population that do not generate their own income. If the dividend is three,for example, there are three members of the productive population for each dependent. Theburden in this case is obviously lighter than if there were only two, enabling more resources to beallocated to constructive production activity.If per-capita income rises as a consequence, a virtuous cycle develops, with stimulus for domesticdemand such as consumer spending, capex, and housing investment so that demand for landincreases as a factor of production, resulting in an increase in land prices. Conversely, if thedemographic dividend shrinks, land prices decline in a reversal of this mechanism.For the remainder of this report we use the “official land price” published by the Ministry of Land, Infrastructure, Transport andTourism in March of each year, which shows prices for January 1. The “benchmark land price” is based on surveys by local authoritiesand shows prices for July 1. The official price is said to have a stronger urban skew and the benchmark a more regional skew, but thetrends are almost the same if the difference in survey timing is discounted. Because the official prices are for January 1, we treat themas data for the year prior to publicatione.g. we treat the 2012 price as 2011 data.高盛全球經(jīng)濟、商品和策略研究42012 年 10 月 19 日日本經(jīng)濟分析The demographic indicator we used in our previous report was the productive population ratio(productive population/total population). The demographic dividend basically moves in the sameway as the productive population ratio. Because its denominator is the dependent populationrather than the total population, which is the sum of the productive population and the dependentpopulation, however, it can capture the above-mentioned mechanism more vividly. In fact, as weshow later, the demographic dividend also has higher explanatory power for land prices.The current trend in the demographic dividend is a halt to the ongoing decline and the beginningsof a rebound (see Exhibit 3, which shows the demographic dividends for the locations covered inExhibit 1). The correlation coefficient with land prices is generally high. We provide more rigorousevidence with our full model later in this report.Exhibit 3: Leveling off/pickup in demographic dividend lies behind the halt in land price declineDemographic dividend(productive population / dependent population)Correlation coefficient for demographic dividendand land prices3.02.72.42.1(x)TokyoTokyoOsakaAichiKanagawaResidentialarea0.610.740.760.74Commercialarea0.580.660.670.61Osaka1.81.5KanagawaAichi19751980198519901995200020052010Source: MIC, MLIT, GS Global ECS Research.Preliminary considerations (2): Prefectural cross sectionobservationsNext we observe the relation between demographics and land price using prefectural cross sections.The average values we obtain for 1970 to 2011 are shown in Exhibit 4. The demographic indicatorswe use are: (1) the productive population ratio (productive population/total population), (2) thedependent population ratio (dependent population/total population), (3) the elderly population ratio(elderly population/total population), and (4) the demographic dividend (productivepopulation/dependent population).There are two points of particular interest. (1) All the demographic indicators show strong linkagewith land prices in the expected direction, but the demographic dividend has the highest explanatorypower measured by coefficient of determination (R2). (2) Demographics strongly influencescommercial land prices, not only residential land prices. In fact, the slopes of the regression lines inthe scatter diagram indicates that demographics has a greater impact on commercial land pricesthan they do on residential land prices.高盛全球經(jīng)濟、商品和策略研究1514151452012 年 10 月 19 日Exhibit 4: Strong link between demographics and land prices can also be found in prefectural cross sections日本經(jīng)濟分析Productive population ratio and land pricesDependent ratio and land prices15(land price; ln)15(land price; ln)1413Residential areaCommercial areay = 20.7x - 1.1R = 0.531413y = -20.7x + 19.6R = 0.53Residential areaCommercial area12121110y = 18.1x - 0.8R = 0.551110y = -18.1x + 17.3R = 0.550.620.640.660.68 0.7 0.72(productive population ratio; x)0.740.250.270.290.310.330.35 0.37 0.39(dependent ratio; x)Elderly ratio and land pricesDemographic dividend and land prices1312(land price; ln)y = -14.9x + 14.8R = 0.34Residential areaCommercial area1312(land price; ln)Residential areaCommercial areay = 2.3x + 8.1R = 0.571110y = -13.7x + 13.2R = 0.381110y = 2.0 x + 7.2R = 0.590.070.090.110.130.150.17 0.19 0.21(elderly ratio; x)1.51.71.92.12.3 2.5 2.7(demographic dividend; x)Source: MIC, MLIT.Our analytical model makes full use of time series and crosssection informationBased on the preliminary observations above, we undertook a rigorous assessment of land pricedeterminants, incorporating all our time series and cross section information in panel analysis. Inchoosing a model, we need to take account of the following three characteristics of land priceformation.(1) Path dependency (stickiness). Unlike financial assets such as stocks and bonds, land cannotbe turned into a standardized product. As a result, its liquidity is very low, and price moves arevery sticky.(2) Investors and developers are nonetheless very sensitive to whether prices are rising orfallingi.e. price momentum. In this context, Japans bubble can be seen as the product of self-fulfilling expectations.(3) There are likely to be regional premiums based on factors such as economic clusters. Japanspolitics and economy are heavily concentrated in the Tokyo area. This may be creating premiumsover and above the impact from productive population inflow.In this report we use a partial adjustment model to take specific account of path dependency(stickiness). In this model, land prices do not adjust to the theoretical level suggested by高盛全球經(jīng)濟、商品和策略研究3362012 年 10 月 19 日日本經(jīng)濟分析fundamentals in any given year. They adjust by only a certain proportion of the differencebetween the theoretical level and the preceding years price (adjustment speed is between 0 and1). We can estimate the adjustment speed together with other parameters.Change in land price = adjustment speed x (that years theoretical land price level previous years land price)In accordance with our earlier findings, we use the demographic dividend as the key determinantof theoretical land prices. To take account of other determinants, we also incorporate prefecturaldata of (1) nominal GDP per head of productive population as a proxy variable for productionactivity and (2) the preceding years change in land price (hereafter price momentum) as anexplanatory variable. The reason for using price momentum is to take account of self-fulfillingexpectations in land price formation.As a macro factor common to all the prefectures, we test significance of the divergence of theMarshallian K from its trend (hereafter Marshallian K). The definition of Marshallian K is moneysupply (M2 + CD)/nominal GDP and deviation from its trend on the upside (downside) denotesexcess (insufficient) liquidity supply for the economy as a whole. We thus have a variable thatshows whether there is too much cash in the economy or not enough.Finally, to capture the level of each prefectures land price premium stemming from factors suchas cluster effect, we added a dummy variable for each prefecture. The dummy variable capturesthe land price divergence for each prefecture (sample period average) when the nationwideaverage is set to zero, which can not be explained by demographics, production activities andland price momentum.Our sample period is 1970 to 2011. (For prefectures where land price data is not available from1970 we used data from the year when data became available .)Results: 95% tracking performanceThe results of our model analysis are summarized in the following four points.(1) Tracking performance is very high, with a 0.95 coefficient of determination for all specifications(see Exhibits 5 and 6). That is to say, our model can explain 95% of Japans prefectural pricemovement over the last 30 years.(2) Adjustment speed is 0.13-0.17 for residential land prices and 0.14-0.15 for commercial landprices, which indicates high path dependency (see Exhibit 5). This signifies that whendemographics and other fundamentals change, only about 15% of the change is reflected in oneyear. Change is incorporated subsequently slowly with lags.(3) The significant variables for theoretical land prices are the demographic dividend, nominal GDPper head of productive population, and land price momentum (see Exhibit 5). Marshallian Ksometimes had a theoretically-correct positive sign in some specifications but it was not statisticallysignificant.(4) As expected, Tokyo has the highest premiums for residential land and commercial land (seeExhibit 7). This is particularly so for commercial land, where the premium is almost three timesthat for second-ranked Osaka. We see the cluster effect at work here, reflecting extremeconcentration in Tokyo. Commercial land premiums show stronger correlation than residentialpremiums with GDP per unit area (see Exhibit 8). Furthermore, Tokyo concentration is the highestglobally (see Exhibit 9). Differences are smaller for residential land, which has high premiums forthe greater Tokyo area (Kanagawa, Saitama, Chiba) and the greater Osaka areas (Kyoto, Hyogo,Osaka).We used dynamic generalized method of moments.高盛全球經(jīng)濟、商品和策略研究72012 年 10 月 19 日日本經(jīng)濟分析Exhibit 5: Combination of demographic dividend and nominal GDP explains land pricesResidential land price modelEstimation equation(1)(2)(3)Population bonus23358.9821985.1810872.17Nominal GDP per personMarshallian K5117.448606.595050.6412537.26Land price momentumAdjustment speedR-square1020.610.170.951010.870.170.95479.570.130.95Commercial land price modelEstimation equation(1)(2)(3)Population bonus98461.53100000.2065730.25Nominal GDP per personMarshallian K20783.30-56812.5717648.14146184.60Land price momentumAdjustment speedR-square7990.350.150.947429.480.150.956370.710.140.95Note: 1. Shaded parts indicate a theoretically-correct sign and statistically significant at 1% level (constant is abbreviated)Note 2. See Exhibit 7 for estimation of land price premium by prefecture.Source: GS Global ECS Research.高盛全球經(jīng)濟、商品和策略研究82012 年 10 月 19 日Exhibit 6: 95% of price movement in Japans 47 prefectures can be explained by our modelExamples of land price estimations (prefectures with top 4 population)日本經(jīng)濟分析Residential land priceCommercial land price1000(Thousand yen/sqm)Tokyo9000(Thousand yen/sqm)Tokyo900800080070060050040030070006000500040003000Commercialarea land priceEstimated2001000Residential area land priceEstimated2000100001975198019851990199520002005201019751980198519901995200020052010500(Thousand yen/sqm)Kanagawa3000(Thousand yen/sqm)Kanagawa450400350300250250020001500Commercialarea land priceEstimated200150100500Residential area land priceEstimated100050001975198019851990199520002005201019751980198519901995200020052010600(Thousand yen/sqm)Osaka6000(Thousand yen/sqm)Osaka5004003002005000400030002000Commercialarea land priceEstimated1000Residential area land priceEstimated100001975198019851990199520002005201019751980198519901995200020052010250(Thousand yen/sqm)Aichi1800(Thousand yen/sqm)Aichi1600200150100140012001000800600Commercial arealand priceEstimated500Residential area land priceEstimated40020001975198019851990199520002005201019751980198519901995200020052010Source: GS Global ECS Research.高盛全球經(jīng)濟、商品和策略研究92012 年 10 月 19 日Exhibit 7: Tokyo has an extremely high price premium particularly for commercial landLand price premium by prefecture日本經(jīng)濟分析HokkaidoAomoriIwateMiyagiResidential land pricesHokkaidoAomoriIwateMiyagiCommercial land pricesAkitaYamagataAkitaYamagatapremiumFukushimaIbaragiTochigiGunmaSaitamapremiumFukushimaIbaragiTochigiGunmaSaitamaChibaTokyoKanagawaNiigataToyamaIshikawaFukuiYamanashiNaganoGif uShizuokaAichiMieShigaKyotoOsakaHyogoNaraWakayamaTottoriShimaneOkayamaHiroshimaYamaguchiTokushimaKagawaEhimeKochiFukuokaSagaNagasakiKumamotoOitaMiyazakiKagoshimaOkinawadiscountChibaTokyoKanagawaNiigataToyamaIshikawaFukuiYamanashiNaganoGif uShizuokaAichiMieShigaKyotoOsakaHyogoNaraWakayamaTottoriShimaneOkayamaHiroshimaYamaguchiTokushimaKagawaEhimeKochiFukuokaSagaNagasakiKumamotoOitaMiyazakiKagoshimaOkinawadiscount-20-100102030-1000100200(thousand yen/sqm)Note: Land price premium measured against national average (zero).Source: GS Global ECS Research.高盛全球經(jīng)濟、商品和策略研究(thousand yen/sqm)301234589102012 年 10 月 19 日Exhibit 8: Concentration of economic activity shows through particularly in commercial land price premiums日本經(jīng)濟分析Residential land price premiums andnominal GDP per unit areaCommercial land price premiums andnominal GDP per unit area(residential land price premium;thousand yen/sqm)250(commercial land price premium;thousand yen/sqm)25y = 0.6451x - 1911.8R = 0.3503200Tokyo2015105KanagawaOsakaTokyo15010050Osakay = 4.5648x - 13068R = 0.62570-5-100-50Kanagawa-150102030 40 50(nominal GDP per unit area;thousand yen/sqm)-1000102030 40 50(nominal GDP per unit area;thousand yen/sqm)Source: Cabinet Office, MIC, MLIT, GS Global ECS ResearchExhibit 9: Tokyo concentration is also high in a global comparisonGlobal nominal GDP rankings (2009, US$ bn)Location of Fortune 500 company headquartersRankCountry / regionUSAJapanChinaGermany(USD bn)13,938.935,035.144,990.533,307.20TokyoBeijingParisNew YorkLondon4844191817France2,631.926710UKItalyTokyo areaBrazilSpain2,180.652,116.631,651.761,622.311,459.42Note: Tokyo area includes Tokyo, Kanagawa, Chiba and Saitama prefectures. 2009 figures are the latest for nominal GDP by prefecture.Source: IMF, Cabinet Office, The Fortune Magazine.GDP per head of productive population is also strongly influencedby the demographic dividendIn our model analysis we found that the demographic dividend and nominal GDP per head ofproductive population are the most fundamental factors in theoretical land price formation. Inaddition, we were interested to find a close link between these two factors. Exhibit 10 shows thedynamic response in each of these variables resulting from a positive shock for the other variable.A positive shock for the demographic dividend produces a significant increase in GDP per head ofproductive population in the third year and the response remains large. However, a positive shockfor nominal GDP per head of productive population produces little significant response from thedemographic dividend. We therefore see no exaggeration in saying that the demographicdivide