[Babacar Mbengue]

Sebastian Kripfganz

A speed-of-adjustment coefficient of about 7 is a clear sign that something is wrong as this would indicate a very explosive process. Unfortunately, it is not possible to identify the reason for this result based on the estimates alone. It might be that the ARDL model is just not suited to fit your data. Do you have any structural breaks in your time series? As a first step towards identifying the problem, you might want to visually compare the fitted values from the regression in levels (without the ec option) with the original data, e.g.:
Code:
ardl lfbcf_priv ltxx lfbcf_pub lide ldebt_ext ldef_fbcf_priv, lags(4 4 3 4 4 4) trendvar(timevar) predict fit, xb twoway (tsline lfbcf_priv) (tsline fit)

Good morning, I run the code above, and that is the result I have

I check also for the Heteroskedasticity of the ardl model and the test of Breusch-Pagan / Cook-Weisberg reports the presence of Heteroskedasticity

estat hettest

Breusch-Pagan / Cook-Weisberg test for heteroskedasticity

Ho: Constant variance

Variables: fitted values of lfbcf_priv


chi2(1) = 5.33

Prob > chi2 = 0.0209



That might be the source of the biaised coefficient?
A another question, how can I apply the test of chow in the ARDL model?

[Sebastian Kripfganz]

A speed-of-adjustment coefficient of about 7 is a clear sign that something is wrong as this would indicate a very explosive process. Unfortunately, it is not possible to identify the reason for this result based on the estimates alone. It might be that the ARDL model is just not suited to fit your data. Do you have any structural breaks in your time series? As a first step towards identifying the problem, you might want to visually compare the fitted values from the regression in levels (without the ec option) with the original data, e.g.:

ardl lfbcf_priv ltxx lfbcf_pub lide ldebt_ext ldef_fbcf_priv, lags(4 4 3 4 4 4) trendvar(timevar)
predict fit, xb
twoway (tsline lfbcf_priv) (tsline fit)

[Babacar Mbengue]

Sebastian Kripfganz

I am not sure I understand your question. What do you mean by the "Pesaran approach" and what do you want to achieve? In other words, what is it that you want to test? If all of your regressors are statistically insignificant, then you already know that there does not exist a long-run relationship between them and your dependent variable.

Sorry, I have some difficulties with the English
I want to use the Pesaran approch of cointegration to determine which variables can affect the private investment. After recent estimation, I find that my lagged dependant variable have a positive sign. According to your previews posts, you said that this coefficient must be in [-1 ; 0]. That's why I asked if I could use the ARDL bounds testing. This is the result I found

[Sebastian Kripfganz]

I use the ardl bounds testing to get the determinants of investment in Senegal. When I use the ardl model to obtain the optimal lag, I have all of my regressors coefficients that are not significant. My dependant variable is TS and after reading Pesaran (2001) I think I fall in a degenerate case. Can I continue to use the Pesaran approch?

I am not sure I understand your question. What do you mean by the "Pesaran approach" and what do you want to achieve? In other words, what is it that you want to test? If all of your regressors are statistically insignificant, then you already know that there does not exist a long-run relationship between them and your dependent variable.

[Babacar Mbengue]

Dear Sebastian Kripfganz ,

I use the ardl bounds testing to get the determinants of investment in Senegal. When I use the ardl model to obtain the optimal lag, I have all of my regressors coefficients that are not significant. My dependant variable is TS and after reading Pesaran (2001) I think I fall in a degenerate case. Can I continue to use the Pesaran approch?

[Sebastian Kripfganz]

sebastian I tried this link net from "http://www.kripfganz.de/stata/" in my STATA-14 but i could not run it. what should i do???

This can happen if you do not have full administrator rights on your PC or if there are some restrictions to Internet access. I have sent you a private message here on Statalist.

[Abdul Rauf]

sebastian I tried this link net from "http://www.kripfganz.de/stata/" in my STATA-14 but i could not run it. what should i do???

[Simon Suleiman]

Thank you very much. I would like also to know whether it is possible to run nonlinear ARDL by using stata ?

[Sebastian Kripfganz]

This error message tells you that you are trying to estimate a large number of models (more than 500) because you are allowing for large orders and probably have a relatively large number of exogenous variables. The error message is implemented as a device to encourage you to think again whether you really want to estimate such large models.

You can increase the allowed number of lag combinations by setting a higher number # with the option maxcombs(#). Depending on the size of your model, the estimations might take a long time. I would recommend in this case to combine this option with the options fast and dots.

Alternatively, you can reduce the actual number of lag combinations by pre-specifying fixed lag numbers for some variables with the lags(numlist) option or by restricting the maximum number of allowed lags for some (or all) variables with the maxlags(numlist) option.

For details about all of these options, please see the ardl help file.

[Simon Suleiman]

Hello everyone, When I run ARDL model the error << r(9) "# of lag permutations (2500) exceeds setting of 'maxcombs' (500)>> appeared. What should I do ?

[Mustapha Immurana]

Thanks so much

[Sebastian Kripfganz]

The best defense is that the estimates are still consistent even if there is no cointegration. In addition, there is of course always the (small) chance that the bounds test incorrectly did not reject the null hypothesis even though it is not true. The ARDL / EC estimates are also robust to the latter situation.

[Mustapha Immurana]

Dear Sebastian,

Thanks so much.

Please my issue has to do with if I am asked why did I still maintain the results since there was no cointegration, how do I defend that?.

[Sebastian Kripfganz]

In all three cases, essentially you cannot reject the null hypothesis of no long-run relationship with the bounds testing procedure. There is nothing wrong with these regressions (provided that your choice of the lag orders is not too conservative). Maybe you want to add a time trend to you regression - see the option trendvar() of ardl - since macroeconomic variables often exhibit a trending behavior.

[Mustapha Immurana]

Dear Sebastian,

The situations below are what I am encountering so please kindly check and see what can be done.

1. Please see the results below where lnsubisdies was stationary at both the levels and after first difference with all other remaining variables being stationary only after first difference and please make suggestions accordingly since the estat btest did not show the existence of co-integration.


ardl lngdp lnrevenueexpenditure lnsubsidies lnloansandadvances dummy, lags(. . . . 0) ec1 regstore(reg3av)

ARDL regression
Model: ec

Sample: 1975 - 2015
Number of obs = 41
Log likelihood = 93.995963
R-squared = .32610714
Adj R-squared = .20718487
Root MSE = .02683852

------------------------------------------------------------------------------
D.lngdp | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
ADJ |
lngdp |
L1. | -.1355375 .0552442 -2.45 0.019 -.2478072 -.0232678
-------------+----------------------------------------------------------------
LR |
lnrevenuee~e |
L1. | .9376934 .1748956 5.36 0.000 .5822628 1.293124
|
lnsubsidies |
L1. | -.0063636 .1545198 -0.04 0.967 -.3203856 .3076584
|
lnloansand~s |
L1. | -.055389 .1027476 -0.54 0.593 -.2641972 .1534192
|
dummy |
L1. | .0850619 .1626433 0.52 0.604 -.245469 .4155927
-------------+----------------------------------------------------------------
SR |
lngdp |
LD. | .4202878 .1416326 2.97 0.005 .1324558 .7081198
|
lnrevenuee~e |
D1. | .1270926 .0651898 1.95 0.060 -.005389 .2595742
|
lnsubsidies |
D1. | -.0008625 .021089 -0.04 0.968 -.0437206 .0419956
|
lnloansand~s |
D1. | -.0075073 .0147952 -0.51 0.615 -.0375748 .0225602
|
dummy |
D1. | .0115291 .0221685 0.52 0.606 -.0335228 .0565809
|
_cons | .4470966 .1504409 2.97 0.005 .141364 .7528292
------------------------------------------------------------------------------

. estat btest

Pesaran/Shin/Smith (2001) ARDL Bounds Test
H0: no levels relationship F = 1.525
t = -2.453

Critical Values (0.1-0.01), F-statistic, Case 3

| [I_0] [I_1] | [I_0] [I_1] | [I_0] [I_1] | [I_0] [I_1]
| L_1 L_1 | L_05 L_05 | L_025 L_025 | L_01 L_01
------+----------------+----------------+----------------+---------------
k_4 | 2.45 3.52 | 2.86 4.01 | 3.25 4.49 | 3.74 5.06
accept if F < critical value for I(0) regressors
reject if F > critical value for I(1) regressors

Critical Values (0.1-0.01), t-statistic, Case 3

| [I_0] [I_1] | [I_0] [I_1] | [I_0] [I_1] | [I_0] [I_1]
| L_1 L_1 | L_05 L_05 | L_025 L_025 | L_01 L_01
------+----------------+----------------+----------------+---------------
k_4 | -2.57 -3.66 | -2.86 -3.99 | -3.13 -4.26 | -3.43 -4.60
accept if t > critical value for I(0) regressors
reject if t < critical value for I(1) regressors

k: # of non-deterministic regressors in long-run relationship
Critical values from Pesaran/Shin/Smith (2001)

2. Also Please see theexamples below where lnGDP (dep. var) and lngrofiscdfct were stationary at both the levels and after first difference with all other remaining variables being stationary only after first difference and please make suggestions accordingly since the estat btest did not show the existence of co-integration.


. ardl lnGDP lngrofiscdfct lngroprmrdfct lnrvnudfct dummy, lags(2 . . 3 0) ec1 regstore(obj5a)

ARDL regression
Model: ec

Sample: 1983 - 2015
Number of obs = 33
Log likelihood = 167.04464
R-squared = .78999092
Adj R-squared = .66398548
Root MSE = .00196845

------------------------------------------------------------------------------
D.lnGDP | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
ADJ |
lnGDP |
L1. | -.0398259 .0302614 -1.32 0.203 -.1029501 .0232983
-------------+----------------------------------------------------------------
LR |
lngrofiscd~t |
L1. | .0719433 .0594311 1.21 0.240 -.0520279 .1959144
|
lngroprmrd~t |
L1. | -.0149252 .0127062 -1.17 0.254 -.0414299 .0115796
|
lnrvnudfct |
L1. | -.0093973 .050849 -0.18 0.855 -.1154664 .0966719
|
dummy |
L1. | .0367571 .0566491 0.65 0.524 -.0814109 .154925
-------------+----------------------------------------------------------------
SR |
lnGDP |
LD. | .4503183 .1627292 2.77 0.012 .1108711 .7897655
|
lngrofiscd~t |
D1. | .0167166 .0070456 2.37 0.028 .0020199 .0314134
LD. | .0062744 .003314 1.89 0.073 -.0006384 .0131872
L2D. | .0100202 .0028935 3.46 0.002 .0039845 .0160559
|
lngroprmrd~t |
D1. | -.0005944 .0006257 -0.95 0.353 -.0018995 .0007107
|
lnrvnudfct |
D1. | -.009855 .0038587 -2.55 0.019 -.0179041 -.001806
LD. | -.0028182 .0017471 -1.61 0.122 -.0064625 .0008261
L2D. | -.0044568 .001266 -3.52 0.002 -.0070977 -.0018159
|
dummy |
D1. | .0014639 .0016406 0.89 0.383 -.0019584 .0048862
|
_cons | .0777555 .0438411 1.77 0.091 -.0136955 .1692065
------------------------------------------------------------------------------

. estat btest

Pesaran/Shin/Smith (2001) ARDL Bounds Test
H0: no levels relationship F = 3.323
t = -1.316

Critical Values (0.1-0.01), F-statistic, Case 3

| [I_0] [I_1] | [I_0] [I_1] | [I_0] [I_1] | [I_0] [I_1]
| L_1 L_1 | L_05 L_05 | L_025 L_025 | L_01 L_01
------+----------------+----------------+----------------+---------------
k_4 | 2.45 3.52 | 2.86 4.01 | 3.25 4.49 | 3.74 5.06
accept if F < critical value for I(0) regressors
reject if F > critical value for I(1) regressors

Critical Values (0.1-0.01), t-statistic, Case 3

| [I_0] [I_1] | [I_0] [I_1] | [I_0] [I_1] | [I_0] [I_1]
| L_1 L_1 | L_05 L_05 | L_025 L_025 | L_01 L_01
------+----------------+----------------+----------------+---------------
k_4 | -2.57 -3.66 | -2.86 -3.99 | -3.13 -4.26 | -3.43 -4.60
accept if t > critical value for I(0) regressors
reject if t < critical value for I(1) regressors

k: # of non-deterministic regressors in long-run relationship
Critical values from Pesaran/Shin/Smith (2001)

3. In this third example lngdp (depvar) and lnntprmr_dfct were stationary at both the levels and after first difference with all other remaining variables being stationary only after first difference but there was no co-integration. Please what do you suggest?

. ardl lnGDP lnntfiscdfct lnntprmr_dfct lnrvnudfct dummy, lags(2 4 4 2 0) ec1 regstore(obj5b)

ARDL regression
Model: ec

Sample: 1982 - 2015
Number of obs = 34
Log likelihood = 170.40245
R-squared = .79608787
Adj R-squared = .60417057
Root MSE = .00227858

------------------------------------------------------------------------------
D.lnGDP | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
ADJ |
lnGDP |
L1. | -.3778128 .1431732 -2.64 0.017 -.6798817 -.0757438
-------------+----------------------------------------------------------------
LR |
lnntfiscdfct |
L1. | .145113 .004913 29.54 0.000 .1347476 .1554784
|
lnntprmr_d~t |
L1. | -.0534484 .0058057 -9.21 0.000 -.0656973 -.0411994
|
lnrvnudfct |
L1. | -.0143404 .0044234 -3.24 0.005 -.0236729 -.0050079
|
dummy |
L1. | -.0004593 .0055216 -0.08 0.935 -.0121089 .0111904
-------------+----------------------------------------------------------------
SR |
lnGDP |
LD. | .785179 .1902565 4.13 0.001 .3837729 1.186585
|
lnntfiscdfct |
D1. | .0088769 .0085029 1.04 0.311 -.0090627 .0268165
LD. | -.0366148 .0130208 -2.81 0.012 -.0640863 -.0091432
L2D. | -.0082798 .0092129 -0.90 0.381 -.0277173 .0111578
L3D. | -.003322 .006488 -0.51 0.615 -.0170105 .0103664
|
lnntprmr_d~t |
D1. | -.0047031 .0028639 -1.64 0.119 -.0107454 .0013392
LD. | .0140574 .0050259 2.80 0.012 .0034536 .0246612
L2D. | .0051476 .0039065 1.32 0.205 -.0030944 .0133896
L3D. | .0024171 .0030758 0.79 0.443 -.0040723 .0089065
|
lnrvnudfct |
D1. | -.0002549 .0022499 -0.11 0.911 -.0050019 .004492
LD. | .0060727 .002081 2.92 0.010 .001682 .0104633
|
dummy |
D1. | -.0001735 .0021155 -0.08 0.936 -.0046369 .0042899
|
_cons | .6215836 .23034 2.70 0.015 .1356087 1.107559
------------------------------------------------------------------------------

. estat btest

Pesaran/Shin/Smith (2001) ARDL Bounds Test
H0: no levels relationship F = 2.086
t = -2.639

Critical Values (0.1-0.01), F-statistic, Case 3

| [I_0] [I_1] | [I_0] [I_1] | [I_0] [I_1] | [I_0] [I_1]
| L_1 L_1 | L_05 L_05 | L_025 L_025 | L_01 L_01
------+----------------+----------------+----------------+---------------
k_4 | 2.45 3.52 | 2.86 4.01 | 3.25 4.49 | 3.74 5.06
accept if F < critical value for I(0) regressors
reject if F > critical value for I(1) regressors

Critical Values (0.1-0.01), t-statistic, Case 3

| [I_0] [I_1] | [I_0] [I_1] | [I_0] [I_1] | [I_0] [I_1]
| L_1 L_1 | L_05 L_05 | L_025 L_025 | L_01 L_01
------+----------------+----------------+----------------+---------------
k_4 | -2.57 -3.66 | -2.86 -3.99 | -3.13 -4.26 | -3.43 -4.60
accept if t > critical value for I(0) regressors
reject if t < critical value for I(1) regressors

k: # of non-deterministic regressors in long-run relationship
Critical values from Pesaran/Shin/Smith (2001)