1. 1 דניאל פלזנשטיין, אייל אשבל וצבי וינוקור אמידה סימולטנית של התנהגות היזם ומחירי קרקע - המקרה של שינויים בשימושי קרקע במטרופולין תל אביב כנס האיגוד הישראלי למדע האזור, אוניברסיטת חיפה, 28.11.10

2. 2 The Motivation In land use models, developer behavior and land prices modeled independently In practice, the two occur simultaneously LU models treat land prices as exogenous. But, developer behavior depends on land prices and vice versa, therefore endogeneity issue. Prices also fixed by expectations of price (rational expectations world)

3. 3 Theory Relative Price Quantity A B D S' (π+1= π) S'' (π+1> π)

4. 4 Supply Z, X = vectors of variables that cause supply/demand curves to shift general price is sum of parcel prices. (–) (+) Demand Equilibriu m

5. 5 Estimation Strategy Maddala (1983): simultaneous equations Use probit two-stage least squares (P2SLS) CDSIMEQ routine (STATA Journal 2003) Land price model (OLS) Developer model (probit)

6. 6 1.Simultaneous equations 2.y * 2 is not observed, rewrite, (1) and (2) as 3.Estimate reduced form 4.Extract predicted values 5.Plug-in fitted values and adjust covariance matrix

7. 7 Estimated Results – Example 1 ln Land Prices Developer Behavior 2 -(-1), Residential – no further development Constant12.43 ** Developer Behavior0.541 * Travel time CBD-0.00253 ** Percent water-0.00710 ** ln resid. units walking dist-0.0808 ** ln resid. units0.104 ** ln distance highway0.0468 ** ln commercial sq. ft.0.0199 ** Mixed Use 1.477 ** Residential-2.377 ** Constant4.113 * ln land prices-0.1300 Access to arterial hwy.-0.5499 * Recent transitions to resid. (walking dist)-0.58853 Recent transitions to same type (walking dist) - 1.4915 ** Percent mixed use (walking dist)0.5465 * Percent same type cells (walking dist)0.01518 * ln resid. units-0.8261 ** -2log likelihood- N2,919 R 2 0.73 LR X 2 - -57.634 238 - 214.5(p<0.000)

8. 8 Estimated Results – Example 2 ln Land Prices Developer Behavior (24-2): Vacant developable – residential (low density) Constant11.56** Developer Behavior0.665** Travel time CBD-0.0066** Percent water-0.0015** ln resid. units walking dist-0.0359* ln resid. units0.0337* Constant-2.766 ln land prices0.026 Recent transitions to resid. (walking dist)0.625* Recent transitions to same type (walking dist)-1.101** Percent residential (walking dist) 0.017 Percent same type cells (walking dist)0.018* ln resid. units0.468** -2log likelihood- N2,696 R 2 0.25 LR X 2 - -40.177 315 - 58.5 (p<0.000) ** p< 0.001; * P<0.05

9. 9 Residential Density (persons per grid cell), 2001-2020

10. 10 Residential Land Values, 2001- 2020

11. 11 Residential Simultaneous estimation predicts more population deconcentration. Residential land values are estimated to be higher in suburban locations than in CBD (using sim. estimation). Indiv. estimation gives opposite picture: higher residential prices closer to CBD: opposite trend.

12. 12 Density of Commercial Development (sq.m.) 2001-2020

13. 13 Non-Residential Land Values, 2001- 2020

14. 14 Non-residential Non-resid sq m: development starts later but reaches more extreme values Similar trends to indiv model estimation. Accentuated suburban non-residential development Simultaneous estimation makes for more extreme values in non- resid land prices. Less smooth price gradient

15. 15 Differences in Households Attributes due to the Two Methods of Estimation Number of Households Average Household Income Δ 2020Δ 2010Δ 2001Δ 2020Δ 2010Δ 2001City Name 5% 1% 0% Ra'anana 2% 0%1%-2%12% Petah Tikva 1% 2% -4%2% Netanya 2% -1% 2%10% Rehovot 1% 0% 2%20% Rishon Leziyon 2% 1% 11%9% Ashdod 1% 3% 1%5% Tel Aviv

16. 16 Differences in Grid Cells Attributes: Estimated Commercial Land Use (sq m) Commercial Land Use (sq.m.) City NameΔ 2001Δ 2010Δ 2020 Ra'anana -18%-4%0% Petah Tikva 27%39%43% Netanya 3%18%20% Rehovot 37%38%37% Rishon Leziyon 25%45%52% Ashdod 31%52%65% Tel Aviv 9%16%15%

17. 17 Differences in Grid Cells Attributes: Number of Estimated Residential Units Residential Units City NameΔ 2001Δ 2010Δ 2020 Ra'anana -2%2%4% Petah Tikva 0%1%3% Netanya 0%1%2% Rehovot -1%0% Rishon Leziyon -2%0% Ashdod 0%1% Tel Aviv 0%1%

18. 18 Differences in Grid Cells Attributes: Change in Share of Residential Land Use Fraction Residential City NameΔ 2001Δ 2010Δ 2020 Ra'anana -23%5% Petah Tikva -9%5% Netanya -6%2% Rehovot -17%-2% Rishon Leziyon -19%-1%-2% Ashdod -8%-3%-4% Tel Aviv 0%1%

19. 19 Conclusions Why is simultaneous estimation more volatile? Technical reason: more noise in estimation due to use of fitted values. No true BLUE estimation- goodness of fit is less robust. But forecasts less likely to be biased; therefore consistently above or below individ. est. (Table). Behavioral focus on land users not land uses. Therefore, endogenity becomes an issue. Past behav and future expectations affect the current. Neighbors behavior- another source of endogeneity.

20. 20 Comparison of Estimated Coefficients for Land Price Model (land conversion from residential to no further development) Estimation Method VariableSimultaneousIndividualΔ Constant 12.43310.9331.500 Travel time CBD -0.002-0.026-0.024 ln resid.units 0.1040.0260.078 ln commercial sq m. 0.0190.0070.012 Mixed use 1.4770.1701.307

21. 21 Actual versus Estimated Population, 2002, 2003, select cities

22. 22 Actual versus Estimated Residential Units, 2002, 2003 select cities

23. 23 Actual versus Estimated Employment 2002, 2003, select cities

24. 24 Actual versus Estimated Commercial Floor Space, 2002, select cities