I am using VAR model and to see the impulse response function I used Cholesky decomposition. The first variable is oil price as a supply shock, second output gap by applying Hodrick-Prescott filter, third for the monetary policy-money base, fourth nominal effective exchange rate, and fifth cpi. After the first difference, they are stationary. The output gap and base money are seasonally adjusted by the census x12 program. residuals are normally and they are not serially correlated .response of CPI to nominal effective exchange rate shock is negative, which means that there is deflation, but at the same time base money has a positive response to neer shock, which means that the national bank uses expansionary policy. and if there is deflation why use national bank expansionary policy? there is something wrong with my model and does anyone has any idea?
Model sounds very impressive!
perhaps have a look at a lower level model for some inspiration
see my old article about IERM
good luck
Fabio Gomez Rodriguez
thanks for your answer. I have already tried it but the result is the same. I cannot understand the logic, why CPI has a negative response when base money has a positive response?
In the short period, it is significant. almost 3 quarter response of CPI is significant, and the response of base money is significant only 2 quarter.
I suppose you worked on quarterly data. You didn't say how many years you used and the order of the VAR model. With 5 variables in you model, you have 25 parameters for each lag so you need very long series to have a high enough degrees of freedom. With 20 years and 2 lags, there remains only 80 - 50 = 30 degrees of freedom. Jan Jansen's paper is here:
Chapter Integrated Exchange Rate Model
Guy Mélard thanks for your response. I have quarterly data from 2003 to 2019. and for cpi, I am using 4 lags. in my model instead of cpi I used PPI with 2 lag and it although gave me a negative response to neer shock. so you think that there is a problem of degrees of freedom?
So the number of observations is 68. You seem to have different lags for each variable so I cannot answer. You can. What is the total number of parameters? It should include the 6x5/2 = 15 error variances and covariances. Is seasonal adjustment really necessary? It can induce perturbations. You can answer privately if you want.
I think you don't have enough observations for this estimation. Moreover, It would be better if you employed SVAR instead of VAR model. The order of variables in the SVAR model is so important. The most exogenous variables should come first and the endogenous variables should come last. I shared the paper that I wrote in 2013 in the following link for your reference.
Article Evaluating the Effects of Monetary Policy Shocks on GCC Countries
Without going into much technicality and assuming that your CPI means Consumer Price Index or Inflation, then your result is a function of your sample/population, timing,variables of your model and model used.
Having said that, if your model is ordinarily CPI= f(Exch, BM, ....et), it can also be the other way round, not minding where causality flows from, then:
1. A negative relationship means exchange rate depreciated (flexible regime) or was devalued (fixed exchange rate regime) within the time period.
2. Positive relationship means exchange rate appreciated or revalued within the time series.
Please read the Purchasing Power Parity (PPP) and The BOP theories of exchange rate.
I do not therefore see any problem here, but a function of your data, time series, variables and modelling. I advise you simple conduct diagnostic tests to confirm stability and global utility of your result.
Dear Irakhi,
The model is okay by me.
But I feel the issue can come from the duration, you need more time frame for a better result.
Then you can also try SVAR with restriction model, that should give you a better result.
Regards
- Badges
- Science topic
- Similar topics
- Control Systems Engineering
- Filtering