You are on page 1of 8

MTB > echo MTB > exec 'a:sugar' Executing from file: a:sugar.

MTB MTB > # MTB > # One way of modelling supply response for an agricultural crop MTB > # is to specify quantity produced or output as the dependent MTB > # variable and to relate this variable to price and other MTB > # relevant supply determining variables. This was the approach MTB > # used previously for modelling wheat supply in a given region MTB > # of Australia. Another approach is to model area (acres) MTB > # sown as the dependent variable. Quantity produced is equal MTB > # to area sown multiplied by yield. Area sown is that component MTB > # of quantity that is completely controlled by the firm. MTB > # The decision to sow a given area will largely depend on the MTB > # firm's desired output at the beginning of the season. Price MTB > # or expected price is a major determinant of desired output. MTB > # After a given area has been sown, it is yield that determines MTB > # quantity produced. Yield will depend on the effects of weather, MTB > # pests and diseases and so is less controllable by the firm. MTB > # Some control on yield can be exerted, however, possibly through MTB > # irrigation, pesticides, weed control, and the like. MTB > # Here we are concerned with modelling the response of area of MTB > # sugarcane sown in a region of Bangladesh. By using area sown, MTB > # not quantity, as the dependent variable, we are eliminating yield MTB > # uncertainty and concentrating on the firm's main decision variable. MTB > # Also, since improvements in technology over time are likely to have MTB > # their effect on yield, not on area, there is no need, as we did for MTB > # wheat, to include a proxy for technological change as an MTB > # explanatory variable. The main product that competes with sugarcane MTB > # is jute. Thus, when farmers decide on an area for sugarcane production MTB > # (A), it is likely to be largely determined by the relative price MTB > # of sugarcane (Ps) to that of jute (Pj). An appropriate economic MTB > # model is A = f(Ps/Pj). Assume a log-linear (constant elasticity) MTB > # functional form. MTB > # Data comprising 34 annual observations on area (thousands of hectares), MTB > # sugarcane price (taka/tonne), and jut price (taka/tonne) are given. MTB > # Example comes from Griffiths et al p 516. MTB > # MTB > # Col 1 Area in thousands of hectares MTB > # Col 2 Price of sugarcane (taka/tonne/ MTB > # Col 3 Price of Jute (taka/tonne) MTB > # MTB > READ 'A:SUGAR' C1-C3

Entering data from file: A:SUGAR.DAT 34 rows read. Row 1 2 3 4 . MTB > MTB > MTB > MTB > MTB > MTB > MTB > MTB > MTB > MTB > MTB > MTB > MTB > MTB > MTB > SUBC> SUBC> SUBC> . . NAME C1 'AREA' C2 'PSUGAR' C3 'PJUTE' # Example application of the Cochrane-Orcutt procedure # taking the model logarea v. logpsugar and logpjute let c4 = loge(c1) let c5 = loge(c2) let c6 = loge(c3) name c4 'lnarea' c5 'lnsug' c6 'lnjute' lag 1 c4 c7 lag 1 c5 c8 lag 1 c6 c9 name c7 'lnarea-1' c8 'lnsug-1' c9 'lnjute-1' # Have lagged the Y and the X's ready for the calculation # of Y* and X*'s Name c10 = 'RESI1' Regress 'lnarea' 2 'lnsug' 'lnjute'; Constant; Residuals 'RESI1'; DW. C1 29 71 42 90 C2 73 108 94 107 C3 970 940 930 970

Regression Analysis The regression equation is lnarea = 3.96 + 0.994 lnsug - 0.685 lnjute Predictor Constant lnsug lnjute s = 0.3956 Coef 3.963 0.9940 -0.6853 Stdev 2.966 0.1424 0.4596 t-ratio 1.34 6.98 -1.49 p 0.191 0.000 0.146

R-sq = 62.1%

R-sq(adj) = 59.6%

Analysis of Variance SOURCE Regression Error Total SOURCE lnsug lnjute DF 2 31 33 DF 1 1 SS 7.9375 4.8525 12.7900 SEQ SS 7.5895 0.3479 MS 3.9687 0.1565 F 25.35 p 0.000

Unusual Observations Obs. lnsug lnarea St.Resid 10 5.16 3.2581 -3.07R 17 5.48 5.5215 X

Fit 4.4390 4.9939

Stdev.Fit 0.0949 0.2362

Residual -1.1809 0.5276 1.66

R denotes an obs. with a large st. resid. X denotes an obs. whose X value gives it large influence. Durbin-Watson statistic = 1.16 MTB > # The DW indicates the presence of possible positive autocorrelation MTB > # other tests should also be carried out to confirm this MTB > # Copy the residuals from column 10 to a scratch column for processing MTB > copy c10 c20 MTB > name c20 'et' MTB > name c21 'et-1' MTB > name c22 'et*et-1' MTB > name c23 'et-1sq' MTB > name c24 'Y*' c25 'Xsug*' c26 'Xjut*' MTB > # Everything is now set up for the calculation of rho MTB > # These steps can be re-used during the iterations MTB > lag 1 c20 c21 MTB > let c22 = c20*c21 MTB > let c23 = c21*c21 MTB > let k1 = sum(c22) MTB > let k2 = sum(c23) MTB > let k3 = k1/k2 MTB > print k3 Data Display K3 0.416959 MTB > # Initial estimate of rho is held in K3 MTB > # Now need to calculate Y* and the X*'s MTB > info Information on the Worksheet

Column C1 C2 C3 C4 C5 C6 C7 C8 C9 C10 C20 C21 C22 C23 C24 C25 C26

Name AREA PSUGAR PJUTE lnarea lnsug lnjute lnarea-1 lnsug-1 lnjute-1 RESI1 et et-1 et*et-1 et-1sq Y* Xsug* Xjut*

Count 34 34 34 34 34 34 34 34 34 34 34 34 34 34 0 0 0

Missing 0 0 0 0 0 0 1 1 1 0 0 1 1 1 0 0 0

Constant Name K1 K2 K3

Value 2.00060 4.79809 0.416959

MTB > let c24 = c4 - k3*c7

MTB > MTB > MTB > MTB > MTB > MTB > SUBC> SUBC> SUBC>

let c25 = c5 - k3*c8 let c26 = c6 - k3*c9 # Now carry out the regression to get the new estimates of # the betas Name c27 = 'RESI2' Regress 'Y*' 2 'Xsug*' 'Xjut*'; Constant; Residuals 'RESI2'; DW.

Regression Analysis

The regression equation is Y* = 3.21 + 1.04 Xsug* - 0.946 Xjut* 33 cases used 1 cases contain missing values Predictor Constant Xsug* Xjut* s = 0.3615 Coef 3.212 1.0438 -0.9459 Stdev 1.420 0.2347 0.3471 t-ratio 2.26 4.45 -2.73 p 0.031 0.000 0.011

R-sq = 42.9%

R-sq(adj) = 39.1%

Analysis of Variance SOURCE Regression Error Total SOURCE Xsug* Xjut* DF 2 30 32 DF 1 1 SS 2.9410 3.9209 6.8620 SEQ SS 1.9704 0.9706 MS 1.4705 0.1307 F 11.25 p 0.000

Unusual Observations Obs. Xsug* Y* St.Resid 10 2.98 1.6996 -2.64R 17 3.19 3.6055 X 18 3.41 2.2086 X

Fit 2.6206 3.2066 2.4291

Stdev.Fit 0.0929 0.1913 0.1918

Residual -0.9209 0.3989 -0.2205 1.30 -0.72

R denotes an obs. with a large st. resid. X denotes an obs. whose X value gives it large influence. Durbin-Watson statistic = 1.84 MTB > # Remember to calculate the new value of the intercept MTB > let k4 = 3.212/(1 - k3) MTB > print k4 Data Display K4 5.50904 MTB > # The estimate of the model is now given by MTB > # MTB > # lnarea = 5.50904 + 1.0438*lnsugar - 0.9459*lnjute

MTB > # MTB > # Can see the DW has improved. Will also need to examine the MTB > # residuals put in c27 to see if they are free of auto. MTB > # In order to see if stability has been reached will need to MTB > # calculate the new set of residuals, re-calculate the value MTB > # of rho and hence Y* and X*'s to see if there is any change MTB > # in the stimates of the betas MTB > # MTB > let c20 = c4 - (5.50904 + 1.0438*c5 - 0.9459*c6) MTB > # MTB > # The new set of residuals are calculated and put into the scratch MTB > # c20. Can now use copy and paste to re-use previous coding. MTB > lag 1 c20 c21 MTB > let c22 = c20*c21 MTB > let c23 = c21*c21 MTB > let k1 = sum(c22) MTB > let k2 = sum(c23) MTB > let k3 = k1/k2 MTB > print k3 Data Display K3 MTB > MTB > MTB > MTB > MTB > MTB > MTB > MTB > MTB > SUBC> SUBC> SUBC> 0.449406 # This is the second estimate of rho. Noe calculate the new # Y* and X*'s let c24 = c4 - k3*c7 let c25 = c5 - k3*c8 let c26 = c6 - k3*c9 # Now carry out the regression to get the new estimates of # the betas Name c27 = 'RESI2' Regress 'Y*' 2 'Xsug*' 'Xjut*'; Constant; Residuals 'RESI2'; DW.

Regression Analysis

The regression equation is Y* = 3.05 + 1.05 Xsug* - 0.959 Xjut* 33 cases used 1 cases contain missing values Predictor Constant Xsug* Xjut* s = 0.3613 Coef 3.051 1.0539 -0.9585 Stdev 1.341 0.2477 0.3410 t-ratio 2.28 4.26 -2.81 p 0.030 0.000 0.009

R-sq = 41.7%

R-sq(adj) = 37.8%

Analysis of Variance SOURCE Regression Error Total SOURCE Xsug* DF 2 30 32 DF 1 SS 2.7978 3.9155 6.7133 SEQ SS 1.7664 MS 1.3989 0.1305 F 10.72 p 0.000

Xjut*

1.0314

Unusual Observations Obs. Xsug* Y* St.Resid 10 2.81 1.5784 -2.57R 17 3.01 3.4564 X 18 3.23 2.0295 X

Fit 2.4762 3.0634 2.2578

Stdev.Fit 0.0935 0.1890 0.1940

Residual -0.8978 0.3930 -0.2283 1.28 -0.75

R denotes an obs. with a large st. resid. X denotes an obs. whose X value gives it large influence. Durbin-Watson statistic = 1.90 MTB > let k4 = MTB > print k4 Data Display K4 MTB MTB MTB MTB MTB MTB MTB MTB 5.54129 let c20 = c4 - ( 5.54129 + 1.0539*c5 lag 1 c20 c21 let c22 = c20*c21 let c23 = c21*c21 let k1 = sum(c22) let k2 = sum(c23) let k3 = k1/k2 print k3 3.051/(1 - k3)

> > > > > > > >

- 0.9585*c6)

Data Display K3 MTB > MTB > MTB > MTB > MTB > MTB > MTB > MTB > SUBC> SUBC> SUBC> 0.451502 # This is the third estimate of rho let c24 = c4 - k3*c7 let c25 = c5 - k3*c8 let c26 = c6 - k3*c9 # Now carry out the regression to get the new estimates of # the betas Name c27 = 'RESI2' Regress 'Y*' 2 'Xsug*' 'Xjut*'; Constant; Residuals 'RESI2'; DW.

Regression Analysis The regression equation is Y* = 3.04 + 1.05 Xsug* - 0.959 Xjut* 33 cases used 1 cases contain missing values Predictor Constant Xsug* Xjut* s = 0.3613 Coef 3.040 1.0546 -0.9593 Stdev 1.336 0.2486 0.3406 t-ratio 2.28 4.24 -2.82 p 0.030 0.000 0.009

R-sq = 41.6%

R-sq(adj) = 37.7%

Analysis of Variance SOURCE Regression Error Total SOURCE Xsug* Xjut* DF 2 30 32 DF 1 1 SS 2.7891 3.9155 6.7046 SEQ SS 1.7537 1.0354 MS 1.3945 0.1305 F 10.68 p 0.000

Unusual Observations Obs. Xsug* Y* St.Resid 10 2.80 1.5705 -2.57R 17 2.99 3.4468 X 18 3.22 2.0179 X

Fit 2.4668 3.0541 2.2467

Stdev.Fit 0.0936 0.1889 0.1942

Residual -0.8963 0.3926 -0.2288 1.27 -0.75

R denotes an obs. with a large st. resid. X denotes an obs. whose X value gives it large influence. Durbin-Watson statistic = 1.90 MTB > let k4 = 3.040/(1 - k3) MTB > print k4 Data Display K4 MTB MTB MTB MTB MTB MTB MTB MTB 5.54241 let c20 = c4 - (5.54241 + 1.0546*c5 lag 1 c20 c21 let c22 = c20*c21 let c23 = c21*c21 let k1 = sum(c22) let k2 = sum(c23) let k3 = k1/k2 print k3

> > > > > > > >

-0.9593*c6)

Data Display K3 0.451605 MTB > # Can now see that rho has converged and hence the estimates of MTB > # the betas. MTB > # You will need to calculate such things as R Square etc MTB > let c99 = c20*c20 MTB > let k5 = sum(c99) MTB > let k6 = 1 - k5/12.79 MTB > # The figure of 12.79 is the Total Sum of Squares from initial regression MTB > print k6 Data Display K6 0.615445 MTB > # This is the actual value of R Square for the corrected model MTB > # Good Luck - I hope this has been a helpful guide A.J. Mabbett

Rev. 0 Feb 97 Rev. 1 March 98

You might also like