Non-linear NK model — high negative initial log posterior despite measurement equations

Hi all,

I am estimating a non-linear New Keynesian model in Dynare and I’m getting what seems like an unusually high negative initial value of the log posterior:

I nitial value of the log posterior (or likelihood): -5899.8732

Data series (from FRED):

  • FEDFUNDS — effective federal funds rate (annual %)
  • GDPC1 — real GDP, billions of chained 2017 USD (quarterly)
  • GDPDEF — GDP deflator (index)
  • CLF16OV — civilian labor force

I have relied on A Guide to Specifying Observation Equations for the
Estimation of DSGE Models. Here are the Transformations and Measurement Equations:

1. Output (y_obs):
Excel:

  • Constructed real GDP per labor force member:
    GDP per capita = LN(GDPC1 / CLF16OV)
  • .mod file:
    y_obs = log(y);

2. Inflation (pi_obs):
Excel:

  • Computed quarterly gross inflation from GDP deflator:
    Inflation = LN(GDPDEF_t / GDPDEF_{t-1})
  • .mod file:
    pi_obs = log(pi);

3. Interest Rate (i_obs):
Excel:

  • Converted the nominal annual % rate to a gross quarterly rate, then took logs:
    r = LN(1 + FEDFUNDS_t / 400)
    .mod file:
    r_obs = log(r);

Despite this, the initial log posterior is very negative, and the posteriors don’t look right.

My question:
Is such a high negative initial log posterior necessarily a problem in a non-linear NK model, or can it be normal depending on the scaling of the data and model?
If it is not normal, what diagnostics or adjustments should I try next?

I have attached both the .mod file and .log files for reference. Please let me know if you need any more files from my end.

Thanks in advance for any guidance.
v3_2_est.log (10.3 KB)
v3_2_est.mod (4.5 KB)

You did not upload the data file. But it looks to me that you did not correctly handle the differing levels of the data and the model variables. The mean inflation rate in the data, for example, will usually not be 0 and the log level of output per capita in the data will be different from log output in the model as well because of different units being used.

1 Like

Thank you so much Prof. for your reply!