0
$\begingroup$

Would it be proper for me to reduce a model by iterating though the coefficients and dropping the ones with high p-values and then refitting and doing this again until all coefficients are significant? The algorithm I have in mind is (starting with the full model):

  1. fit regression on current set of regressors
  2. find coefficient with highest p-value
  3. drop regressor variable from model
  4. go to step 1

I select the "highest p-value" because the null hypothesis in my software package (python statsmodels) is that the coefficient is 0 so we only keep ones with low p-values. I can potentially then perform a stepAIC after running the above step. For example, here is a model I ran just now which I'd like to reduce.

                            OLS Regression Results                            
==============================================================================
Dep. Variable:                  c0000   R-squared:                       0.183
Model:                            OLS   Adj. R-squared:                  0.105
Method:                 Least Squares   F-statistic:                     2.337
Date:                Fri, 14 May 2021   Prob (F-statistic):           3.53e-28
Time:                        13:14:07   Log-Likelihood:                 23470.
No. Observations:                3285   AIC:                        -4.636e+04
Df Residuals:                    2997   BIC:                        -4.461e+04
Df Model:                         287                                         
Covariance Type:            nonrobust                                         
==============================================================================
                 coef    std err          t      P>|t|      [0.025      0.975]
------------------------------------------------------------------------------
Intercept  -1.323e-05   3.86e-06     -3.428      0.001   -2.08e-05   -5.66e-06
c1620         -0.0211      0.014     -1.541      0.123      -0.048       0.006
c1655          0.0021      0.015      0.137      0.891      -0.027       0.032
c0705          0.0300      0.010      2.890      0.004       0.010       0.050
c1210          0.0095      0.014      0.671      0.503      -0.018       0.037
c1900         -0.0082      0.013     -0.634      0.526      -0.034       0.017
c0905          0.0034      0.012      0.280      0.780      -0.021       0.027
c0650          0.0064      0.015      0.424      0.672      -0.023       0.036
c1130          0.0368      0.015      2.400      0.016       0.007       0.067
c1015          0.0477      0.014      3.426      0.001       0.020       0.075
c0300          0.0313      0.023      1.346      0.178      -0.014       0.077
c1535          0.0084      0.013      0.671      0.502      -0.016       0.033
c1650          0.0301      0.016      1.862      0.063      -0.002       0.062
c0235         -0.0446      0.024     -1.848      0.065      -0.092       0.003
c0545          0.0030      0.025      0.120      0.904      -0.046       0.052
c1515          0.0264      0.011      2.331      0.020       0.004       0.049
c2240         -0.0799      0.029     -2.763      0.006      -0.137      -0.023
c1255         -0.0166      0.010     -1.612      0.107      -0.037       0.004
c1820          0.0142      0.017      0.843      0.399      -0.019       0.047
c1215          0.0297      0.012      2.573      0.010       0.007       0.052
c2340         -0.0208      0.017     -1.200      0.230      -0.055       0.013
c2155         -0.0734      0.029     -2.491      0.013      -0.131      -0.016
c2145          0.0401      0.033      1.224      0.221      -0.024       0.104
c0920          0.0165      0.014      1.165      0.244      -0.011       0.044
c0215          0.0178      0.024      0.749      0.454      -0.029       0.064
c1855         -0.0314      0.018     -1.736      0.083      -0.067       0.004
c0535          0.0373      0.025      1.502      0.133      -0.011       0.086
c1310          0.0127      0.011      1.160      0.246      -0.009       0.034
c1420         -0.0342      0.010     -3.307      0.001      -0.054      -0.014
c1705         -0.0188      0.014     -1.318      0.188      -0.047       0.009
c1005         -0.0167      0.014     -1.212      0.225      -0.044       0.010
c0810          0.0198      0.011      1.760      0.079      -0.002       0.042
c0620          0.0100      0.018      0.568      0.570      -0.024       0.044
c0255         -0.0397      0.025     -1.617      0.106      -0.088       0.008
c2320          0.0530      0.027      1.959      0.050   -4.49e-05       0.106
c2200         -0.0014      0.009     -0.159      0.874      -0.019       0.016
c0520          0.0447      0.026      1.696      0.090      -0.007       0.096
c0835          0.0020      0.012      0.170      0.865      -0.021       0.025
c0155         -0.0242      0.023     -1.055      0.292      -0.069       0.021
c1000          0.0033      0.012      0.270      0.787      -0.021       0.027
c2040         -0.0329      0.026     -1.270      0.204      -0.084       0.018
c0530         -0.0024      0.026     -0.094      0.925      -0.053       0.049
c0250         -0.0072      0.024     -0.296      0.767      -0.054       0.040
c1240         -0.0170      0.008     -2.055      0.040      -0.033      -0.001
c1430         -0.0120      0.011     -1.102      0.270      -0.033       0.009
c0610          0.0213      0.015      1.373      0.170      -0.009       0.052
c1050          0.0436      0.016      2.762      0.006       0.013       0.075
c1320         -0.0146      0.012     -1.255      0.210      -0.037       0.008
c1710         -0.0257      0.016     -1.600      0.110      -0.057       0.006
c0855          0.0050      0.014      0.354      0.723      -0.023       0.033
c0130         -0.0459      0.022     -2.123      0.034      -0.088      -0.004
c0350          0.0087      0.028      0.316      0.752      -0.045       0.063
c0210          0.0229      0.025      0.909      0.364      -0.026       0.072
c1020         -0.0027      0.015     -0.184      0.854      -0.032       0.026
c0625         -0.0370      0.017     -2.219      0.027      -0.070      -0.004
c1915         -0.0137      0.019     -0.729      0.466      -0.051       0.023
c0600          0.0142      0.017      0.832      0.406      -0.019       0.048
c2135         -0.0091      0.031     -0.292      0.770      -0.070       0.052
c0305         -0.0187      0.024     -0.789      0.430      -0.065       0.028
c1330          0.0188      0.008      2.503      0.012       0.004       0.034
c1545          0.0085      0.012      0.692      0.489      -0.016       0.033
c2330          0.0163      0.028      0.582      0.560      -0.038       0.071
c1530         -0.0189      0.012     -1.603      0.109      -0.042       0.004
c1640         -0.0386      0.015     -2.512      0.012      -0.069      -0.008
c1450         -0.0291      0.011     -2.697      0.007      -0.050      -0.008
c0200         -0.0389      0.022     -1.795      0.073      -0.081       0.004
c0310          0.0906      0.026      3.464      0.001       0.039       0.142
c1815          0.0201      0.015      1.367      0.172      -0.009       0.049
c1315          0.0050      0.010      0.478      0.633      -0.015       0.025
c0120         -0.0145      0.021     -0.700      0.484      -0.055       0.026
c0605          0.0068      0.015      0.455      0.649      -0.022       0.036
c1630          0.0136      0.014      0.953      0.341      -0.014       0.042
c0040          0.0090      0.023      0.387      0.699      -0.037       0.055
c0435         -0.0921      0.029     -3.172      0.002      -0.149      -0.035
c0755         -0.0116      0.013     -0.886      0.376      -0.037       0.014
c0555          0.0050      0.023      0.221      0.825      -0.039       0.049
c2125          0.0148      0.031      0.483      0.629      -0.045       0.075
c0345         -0.0166      0.030     -0.551      0.581      -0.075       0.042
c0800          0.0244      0.011      2.208      0.027       0.003       0.046
c1610          0.0076      0.013      0.592      0.554      -0.018       0.033
c0455          0.0385      0.032      1.211      0.226      -0.024       0.101
c0410         -0.0141      0.024     -0.583      0.560      -0.062       0.033
c1910          0.0206      0.016      1.255      0.210      -0.012       0.053
c0025         -0.0155      0.022     -0.705      0.481      -0.059       0.028
c0735          0.0283      0.013      2.240      0.025       0.004       0.053
c0510          0.0655      0.026      2.545      0.011       0.015       0.116
c0335         -0.0153      0.027     -0.559      0.576      -0.069       0.038
c0240         -0.0022      0.025     -0.088      0.930      -0.051       0.046
c1200         -0.0295      0.013     -2.328      0.020      -0.054      -0.005
c1335          0.0090      0.007      1.274      0.203      -0.005       0.023
c0840         -0.0142      0.013     -1.108      0.268      -0.039       0.011
c0645         -0.0089      0.017     -0.531      0.595      -0.042       0.024
c0330          0.0298      0.030      0.993      0.321      -0.029       0.089
c1925          0.0145      0.019      0.746      0.456      -0.024       0.052
c1930         -0.0003      0.020     -0.017      0.986      -0.039       0.038
c0830         -0.0073      0.012     -0.604      0.546      -0.031       0.016
c0205          0.0291      0.023      1.266      0.206      -0.016       0.074
c1600         -0.0064      0.010     -0.649      0.517      -0.026       0.013
c0740          0.0228      0.013      1.768      0.077      -0.002       0.048
c1250          0.0183      0.010      1.910      0.056      -0.000       0.037
c2100      -4.595e-05      0.007     -0.007      0.994      -0.013       0.013
c1030          0.0035      0.015      0.228      0.819      -0.026       0.033
c1730         -0.0291      0.017     -1.716      0.086      -0.062       0.004
c2255          0.0331      0.027      1.204      0.229      -0.021       0.087
c2105          0.0025      0.009      0.278      0.781      -0.015       0.020
c1225          0.0117      0.014      0.826      0.409      -0.016       0.040
c1145         -0.0005      0.013     -0.039      0.969      -0.026       0.025
c1745         -0.0355      0.017     -2.116      0.034      -0.068      -0.003
c1750          0.0074      0.017      0.424      0.672      -0.027       0.042
c1055          0.0154      0.016      0.965      0.335      -0.016       0.047
c2220         -0.0140      0.030     -0.469      0.639      -0.073       0.045
c1635         -0.0013      0.015     -0.085      0.933      -0.031       0.028
c1135         -0.0317      0.015     -2.093      0.036      -0.061      -0.002
c1440         -0.0157      0.011     -1.408      0.159      -0.038       0.006
c0640         -0.0008      0.016     -0.052      0.959      -0.033       0.031
c2315         -0.0387      0.027     -1.432      0.152      -0.092       0.014
c1345          0.0115      0.009      1.290      0.197      -0.006       0.029
c0825          0.0360      0.013      2.762      0.006       0.010       0.062
c0420         -0.0588      0.033     -1.792      0.073      -0.123       0.006
c1615         -0.0406      0.014     -2.948      0.003      -0.068      -0.014
c2115          0.0190      0.028      0.686      0.493      -0.035       0.073
c2335          0.0278      0.027      1.034      0.301      -0.025       0.080
c0100          0.0343      0.022      1.560      0.119      -0.009       0.077
c1700          0.0195      0.014      1.376      0.169      -0.008       0.047
c0750          0.0329      0.014      2.426      0.015       0.006       0.060
c0925         -0.0223      0.014     -1.541      0.123      -0.051       0.006
c1555         -0.0116      0.013     -0.916      0.360      -0.036       0.013
c1040          0.0288      0.015      1.885      0.060      -0.001       0.059
c1455         -0.0111      0.011     -1.053      0.292      -0.032       0.010
c0930         -0.0047      0.014     -0.334      0.738      -0.032       0.023
c1120          0.0065      0.014      0.448      0.654      -0.022       0.035
c2015         -0.0532      0.024     -2.177      0.030      -0.101      -0.005
c1150          0.0031      0.012      0.267      0.789      -0.020       0.026
c2110          0.0138      0.026      0.543      0.588      -0.036       0.064
c2010          0.0134      0.023      0.580      0.562      -0.032       0.059
c0245         -0.0055      0.026     -0.217      0.828      -0.056       0.044
c0230          0.0355      0.025      1.407      0.160      -0.014       0.085
c0700          0.0058      0.012      0.492      0.623      -0.017       0.029
c0505          0.0117      0.025      0.459      0.646      -0.038       0.062
c1010         -0.0173      0.015     -1.181      0.238      -0.046       0.011
c1845          0.0503      0.020      2.579      0.010       0.012       0.089
c1940          0.0031      0.020      0.152      0.880      -0.037       0.043
c1725          0.0270      0.016      1.639      0.101      -0.005       0.059
c1140         -0.0145      0.015     -0.964      0.335      -0.044       0.015
c2225          0.0234      0.028      0.838      0.402      -0.031       0.078
c1245          0.0043      0.009      0.504      0.615      -0.012       0.021
c1850         -0.0157      0.020     -0.796      0.426      -0.054       0.023
c0745          0.0273      0.013      2.117      0.034       0.002       0.053
c1950          0.0076      0.021      0.358      0.720      -0.034       0.049
c1445         -0.0271      0.011     -2.570      0.010      -0.048      -0.006
c0050          0.0164      0.023      0.709      0.478      -0.029       0.062
c2350         -0.0219      0.026     -0.858      0.391      -0.072       0.028
c0030          0.0427      0.024      1.781      0.075      -0.004       0.090
c0440          0.0468      0.030      1.557      0.120      -0.012       0.106
c0710         -0.0056      0.011     -0.492      0.622      -0.028       0.017
c2230         -0.0200      0.031     -0.652      0.515      -0.080       0.040
c2345          0.0007      0.024      0.030      0.976      -0.047       0.049
c0515          0.0410      0.028      1.461      0.144      -0.014       0.096
c2310         -0.0475      0.030     -1.590      0.112      -0.106       0.011
c1840          0.0240      0.016      1.465      0.143      -0.008       0.056
c0935          0.0352      0.013      2.806      0.005       0.011       0.060
c0425         -0.0794      0.033     -2.395      0.017      -0.144      -0.014
c0320         -0.0263      0.028     -0.943      0.346      -0.081       0.028
c0055         -0.0163      0.019     -0.858      0.391      -0.053       0.021
c2250         -0.0565      0.029     -1.975      0.048      -0.113      -0.000
c0945          0.0030      0.014      0.209      0.835      -0.025       0.031
c0525         -0.0607      0.028     -2.184      0.029      -0.115      -0.006
c0140         -0.0044      0.022     -0.206      0.837      -0.047       0.038
c1715         -0.0266      0.015     -1.735      0.083      -0.057       0.003
c2005         -0.0076      0.017     -0.447      0.655      -0.041       0.026
c1415          0.0072      0.010      0.732      0.464      -0.012       0.027
c1735         -0.0173      0.015     -1.185      0.236      -0.046       0.011
c2025          0.0092      0.026      0.355      0.723      -0.042       0.060
c1830          0.0310      0.018      1.713      0.087      -0.004       0.066
c2235          0.0037      0.030      0.126      0.900      -0.054       0.062
c0010         -0.0042      0.019     -0.219      0.827      -0.042       0.034
c0635         -0.0338      0.016     -2.105      0.035      -0.065      -0.002
c1520         -0.0172      0.012     -1.391      0.164      -0.041       0.007
c1350         -0.0142      0.010     -1.464      0.143      -0.033       0.005
c0035         -0.0155      0.021     -0.727      0.467      -0.057       0.026
c0445          0.0249      0.032      0.777      0.437      -0.038       0.088
c0225         -0.0239      0.024     -0.995      0.320      -0.071       0.023
c0910         -0.0190      0.013     -1.442      0.149      -0.045       0.007
c1755          0.0225      0.018      1.260      0.208      -0.013       0.058
c1435         -0.0201      0.010     -1.930      0.054      -0.041       0.000
c0340         -0.0885      0.026     -3.371      0.001      -0.140      -0.037
c0655          0.0246      0.015      1.648      0.099      -0.005       0.054
c0315         -0.0327      0.029     -1.116      0.265      -0.090       0.025
c0815         -0.0066      0.012     -0.531      0.596      -0.031       0.018
c1100          0.0208      0.014      1.464      0.143      -0.007       0.049
c0955         -0.0159      0.015     -1.050      0.294      -0.045       0.014
c2035          0.0469      0.026      1.825      0.068      -0.003       0.097
c0500          0.0040      0.027      0.149      0.881      -0.048       0.056
c0115         -0.0138      0.021     -0.653      0.514      -0.055       0.028
c1550          0.0007      0.013      0.056      0.955      -0.025       0.027
c1115         -0.0108      0.015     -0.717      0.473      -0.040       0.019
c1305          0.0070      0.010      0.706      0.480      -0.012       0.026
c0015          0.0275      0.021      1.335      0.182      -0.013       0.068
c0045         -0.0014      0.022     -0.063      0.950      -0.044       0.041
c0430         -0.1043      0.033     -3.150      0.002      -0.169      -0.039
c1105         -0.0387      0.015     -2.555      0.011      -0.068      -0.009
c0020         -0.0524      0.023     -2.314      0.021      -0.097      -0.008
c1155          0.0168      0.015      1.152      0.249      -0.012       0.045
c0400         -0.0263      0.028     -0.938      0.348      -0.081       0.029
c1800         -0.0096      0.010     -0.963      0.335      -0.029       0.010
c0105         -0.0097      0.020     -0.479      0.632      -0.050       0.030
c2055         -0.0240      0.025     -0.955      0.340      -0.073       0.025
c0450         -0.0150      0.031     -0.481      0.630      -0.076       0.046
c1825         -0.0108      0.018     -0.605      0.545      -0.046       0.024
c1300         -0.0006      0.010     -0.054      0.957      -0.021       0.020
c0540         -0.0524      0.026     -2.018      0.044      -0.103      -0.001
c1205         -0.0312      0.013     -2.357      0.019      -0.057      -0.005
c0725         -0.0408      0.013     -3.100      0.002      -0.067      -0.015
c0145          0.0541      0.022      2.482      0.013       0.011       0.097
c1125         -0.0061      0.016     -0.385      0.700      -0.037       0.025
c1540          0.0123      0.012      1.042      0.297      -0.011       0.036
c1220          0.0113      0.012      0.958      0.338      -0.012       0.034
c1230         -0.0026      0.006     -0.433      0.665      -0.014       0.009
c1510          0.0094      0.010      0.902      0.367      -0.011       0.030
c0415          0.0412      0.030      1.371      0.170      -0.018       0.100
c0355         -0.0218      0.030     -0.731      0.465      -0.080       0.037
c2150         -0.0052      0.030     -0.174      0.862      -0.063       0.053
c1400         -0.0023      0.008     -0.274      0.784      -0.019       0.014
c1805         -0.0149      0.011     -1.369      0.171      -0.036       0.006
c1945         -0.0205      0.020     -1.033      0.302      -0.059       0.018
c2045          0.0530      0.024      2.201      0.028       0.006       0.100
c1235         -0.0117      0.008     -1.415      0.157      -0.028       0.005
c0615          0.0133      0.017      0.778      0.437      -0.020       0.047
c2030         -0.0444      0.024     -1.822      0.069      -0.092       0.003
c1920          0.0375      0.019      2.000      0.046       0.001       0.074
c2050          0.0954      0.025      3.849      0.000       0.047       0.144
c0005          0.0059      0.015      0.399      0.690      -0.023       0.035
c2130          0.0307      0.030      1.029      0.304      -0.028       0.089
c0940          0.0242      0.014      1.679      0.093      -0.004       0.052
c2205          0.0187      0.021      0.870      0.384      -0.023       0.061
c1405         -0.0193      0.009     -2.254      0.024      -0.036      -0.003
c2305          0.0333      0.024      1.364      0.173      -0.015       0.081
c1935          0.0058      0.019      0.301      0.764      -0.032       0.044
c0325         -0.0593      0.030     -1.969      0.049      -0.118      -0.000
c1810         -0.0213      0.014     -1.497      0.134      -0.049       0.007
c1835          0.0327      0.015      2.130      0.033       0.003       0.063
c0900          0.0037      0.011      0.325      0.745      -0.019       0.026
c1425         -0.0215      0.011     -1.965      0.049      -0.043   -5.06e-05
c0125         -0.0408      0.022     -1.891      0.059      -0.083       0.002
c0850         -0.0030      0.013     -0.236      0.814      -0.028       0.022
c1325         -0.0477      0.012     -3.888      0.000      -0.072      -0.024
c1525          0.0098      0.012      0.821      0.412      -0.014       0.033
c0135          0.0353      0.020      1.731      0.084      -0.005       0.075
c1505          0.0205      0.010      2.140      0.032       0.002       0.039
c1740          0.0008      0.018      0.044      0.965      -0.034       0.035
c0805          0.0098      0.011      0.924      0.355      -0.011       0.031
c2020         -0.0237      0.024     -0.981      0.327      -0.071       0.024
c2120         -0.0276      0.028     -1.002      0.316      -0.082       0.026
c0950         -0.0076      0.015     -0.517      0.605      -0.037       0.021
c1605         -0.0037      0.013     -0.288      0.773      -0.029       0.021
c1025         -0.0109      0.014     -0.755      0.450      -0.039       0.017
c1355          0.0063      0.011      0.588      0.556      -0.015       0.027
c0845          0.0061      0.013      0.469      0.639      -0.019       0.032
c1905         -0.0246      0.014     -1.774      0.076      -0.052       0.003
c2000         -0.0022      0.018     -0.122      0.903      -0.038       0.033
c1955         -0.0034      0.021     -0.160      0.873      -0.045       0.038
c0150         -0.0024      0.022     -0.110      0.912      -0.045       0.040
c0110         -0.0321      0.021     -1.541      0.124      -0.073       0.009
c2210         -0.0170      0.024     -0.724      0.469      -0.063       0.029
c1340         -0.0003      0.009     -0.029      0.977      -0.018       0.017
c0220          0.0204      0.026      0.789      0.430      -0.030       0.071
c1110          0.0089      0.015      0.609      0.542      -0.020       0.038
c0405         -0.0499      0.030     -1.662      0.097      -0.109       0.009
c0915         -0.0351      0.013     -2.605      0.009      -0.062      -0.009
c2140          0.0210      0.030      0.690      0.490      -0.039       0.081
c0630          0.0278      0.016      1.718      0.086      -0.004       0.059
c2355          0.0119      0.023      0.512      0.609      -0.034       0.057
c0720          0.0248      0.013      1.939      0.053      -0.000       0.050
c1410          0.0022      0.009      0.237      0.813      -0.016       0.021
c1645         -0.0095      0.016     -0.604      0.546      -0.040       0.021
c1500          0.0441      0.009      5.129      0.000       0.027       0.061
c2325          0.0680      0.029      2.338      0.019       0.011       0.125
c0730          0.0118      0.011      1.060      0.289      -0.010       0.034
c0715         -0.0039      0.012     -0.317      0.751      -0.028       0.020
c1045         -0.0221      0.016     -1.411      0.158      -0.053       0.009
c1625         -0.0135      0.014     -0.952      0.341      -0.041       0.014
c0820          0.0259      0.012      2.146      0.032       0.002       0.050
c1720          0.0090      0.016      0.553      0.580      -0.023       0.041
c2245         -0.0404      0.030     -1.346      0.178      -0.099       0.018
c2300          0.0118      0.024      0.486      0.627      -0.036       0.060
c2215          0.0115      0.026      0.435      0.664      -0.040       0.063
c1035         -0.0063      0.015     -0.423      0.672      -0.036       0.023
c0550          0.0230      0.024      0.942      0.346      -0.025       0.071
==============================================================================
Omnibus:                     4043.713   Durbin-Watson:                   1.954
Prob(Omnibus):                  0.000   Jarque-Bera (JB):          4176392.615
Skew:                          -5.781   Prob(JB):                         0.00
Kurtosis:                     177.295   Cond. No.                     1.12e+04
==============================================================================

Notes:
[1] Standard Errors assume that the covariance matrix of the errors is correctly specified.
[2] The condition number is large, 1.12e+04. This might indicate that there are
strong multicollinearity or other numerical problems.

MSr: 0.0003056669174634134

MSe: 0.00019993423813406298
$\endgroup$
8
  • 1
    $\begingroup$ This is a step-wise procedure that used to be done some. However, if your goal is interpreting your coefficients then this will result in a bad model. If your goal is to predict then this will result in a bad model. what exactly is your goal? $\endgroup$ – Tylerr May 14 at 12:46
  • $\begingroup$ @Tylerr: I agree with asking for the goal and with your statement that backward selection based on p-values is somewhat outdated and has problems. However there are situations in which it will actually work well, so in general you can't say "this will result in a bad model" - it may, but then it may not. $\endgroup$ – Lewian May 14 at 13:11
  • 1
    $\begingroup$ @Lewian, of course there are situations where it can work just as it is possible to flip a coin and get the best model. I think it is much more prudent to recommend, whole-heartedly, better methods for variable selection or regularization. This method has been pretty beat up since the 90s, I am surprised my statement is controversial. $\endgroup$ – Tylerr May 14 at 13:30
  • $\begingroup$ @Tylerr: What is controversial or rather in fact wrong is the overgeneralisation in your first comment. In order to recommend method X and say that method Y has problems you don't need to pretend that method Y gets it wrong all the time. In fact it is not too difficult to find instances in which it outperforms lasso, for example. I teach variable selection and wanted to demonstrate with a few examples that lasso does better than backward selection, but backward selection won the first four instances that I tried out. $\endgroup$ – Lewian May 14 at 13:40
  • $\begingroup$ We shouldn't pretend the statisticians who recommended this in the sixties and used it in the seventies were idiots. $\endgroup$ – Lewian May 14 at 13:42
1
$\begingroup$

The short answer is "No".

  • Stepwise regression is popular, but problematic. Frank Harrell lists several reasons why stepwise is problematic in his book Regression Modelling Strategies.

  • If your goal is interpretability, the procedure is not guaranteed to pick important variables (so you can't say for instance that a variable with p>0.05 is unimportant). Moreover, the p value tells you nothing about the effect of the variable. It is completely possible that you have a variable with a large effect but also large uncertainty. I've demonstrated that here and elsewhere.

  • If your goal is prediction, variable selection can often hurt performance as I show here.

  • If your goal is inference, you're out of luck again because the p values lose all meaning since they do not condition on the selection procedure.

Can you tell us what the goal of your analysis is? That can help us choose a better way of moving forward.

$\endgroup$
2
  • $\begingroup$ Thanks for the answer. The regression you see in my question is for EUR/USD 5-min returns split into the time of day series. So for example 16:35 will have its own daily series. My MSc professor suggested I do a regression to spot patterns. I'm unlikely to see patterns in EUR/USD but I'm trying to understand the technique and apply it to other assets. $\endgroup$ – s5s May 14 at 16:01
  • $\begingroup$ @s5s This doesn't sound like an appropriate method for OLS. $\endgroup$ – Demetri Pananos May 14 at 16:43

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.