Short run coef.

[nipnipj]

Could anyone, please, confirm the following. I think the output is incorrect, the estimated coefficient of TBILL1M belongs to DUM0708 and vice versa.

multipliers(uecm, type = "sr")
         Term     Estimate Std. Error    t value     Pr(>|t|)
1 (Intercept) -0.004117161 0.01886963 -0.2181898 8.275179e-01
2     TBILL3M  0.014984312 0.01533846  0.9769109 3.298686e-01
3     TBILL1M  1.274725456 0.07200374 17.7035995 6.562244e-42
4     DUM0708 -0.426651174 0.07189549 -5.9343243 1.394578e-08

[Natsiopoulos]

I can check it if you post the results of the UECM model.

[nipnipj]

> summary(uecm) 

Time series regression with "ts" data:
Start = 2, End = 196

Call:
dynlm::dynlm(formula = full_formula, data = data, start = start, 
    end = end)

Residuals:
      Min        1Q    Median        3Q       Max 
-0.271644 -0.030401 -0.002058  0.027445  0.243267 

Coefficients:
               Estimate Std. Error t value Pr(>|t|)    
(Intercept)   -0.004117   0.018870  -0.218    0.828    
L(TBILL6M, 1) -0.554607   0.066150  -8.384 1.18e-14 ***
L(TBILL3M, 1)  1.111351   0.126289   8.800 8.80e-16 ***
L(TBILL1M, 1) -0.555433   0.075213  -7.385 4.82e-12 ***
DUM0708        0.014984   0.015338   0.977    0.330    
d(TBILL3M)     1.274725   0.072004  17.704  < 2e-16 ***
d(TBILL1M)    -0.426651   0.071895  -5.934 1.39e-08 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 0.06478 on 188 degrees of freedom
Multiple R-squared:  0.8828,	Adjusted R-squared:  0.8791 
F-statistic: 236.1 on 6 and 188 DF,  p-value: < 2.2e-16

[Natsiopoulos]

Indeed! This is strange. Thank you for raising this issue!
Could you provide a reproducible example (data and code)?
I am afraid this is a serious mistake that I have to fix!

[nipnipj]

Sure.
bby.csv

model <- auto_ardl(
  TBILL6M ~ TBILL3M + TBILL1M + DUM0708,
  data = data, selection = "AIC",
  max_order = c(6,6,6,0),
  starting_order = c(1,0,0,0))

[Natsiopoulos]

Thank you.

So, some early finding:
If you run multipliers(ardl, type = "sr") instead of multipliers(uecm, type = "sr") it produces the correct estimations.
I am on it to find why it is messed up when you use uecm.

Regardless of this issue, you should probably use:
model <- auto_ardl( TBILL6M ~ TBILL3M + TBILL1M | DUM0708, data = data, selection = "AIC", max_order = c(6,6,6), starting_order = c(1,0,0))
as DUM0708 is a dummy variable. This may be part of the problem (but it shouldn't).

[nipnipj]

I've tested the package with another dataset. ARDL, CEC, ECM, SR & LR coefficients, F-statistic bound test, t-statistic bound test, look fine. Sample size is 47, tho.

[Natsiopoulos]

The problem arises when there is a 0 order. It messes the order of the rows (fortunately, not a methodological mistake, just a row.names issue). I am fixing it right now. I will keep you updated for more details tomorrow.