Solved – Jags error with dgamma

bayesianjagsmarkov-chain-montecarlo

I am trying to sample from a Gamma distribution in JAGS

gd[i] ~ dgamma(k,r)

where k and r are have priors such that they are positive. However I get the following error when I compile the model:

  Error in node gd[9]
Unobserved node inconsistent with unobserved parents at initialization

Any ideas?

EDIT: BACKGROUND

What I am trying to do is to estimate the parameters of the CIR process with Bayesian techniques. In order to assess the estimator, I have simulated some data with given parameters on a quarterly basis (timeStep=90):

jdata <- list(timeSerie=c(35000, 47123.3999346341, 55373.904244638, 71755.7487795017, 91246.3370619453, 76969.2358694695, 83390.030308266, 80657.1222156083, 76784.466470129, 83962.4642428193, 85335.5796326587, 87018.8279086056, 74059.9912730579, 69683.5404626939, 90356.0122717258, 87240.7793078735, 74276.2931903815, 39899.5948334073, 37911.738710607, 73811.7536157058, 69334.5390010963, 51326.297331755, 42711.7358571988, 32409.9939465308, 16663.7353817099, 22421.1721803643, 38604.0551895535, 50181.3269231991, 44346.3517857709, 52127.2156791002, 54740.1181745155, 41869.456839522, 50665.5626313423, 51121.4339441224, 39059.432945341, 44350.0205383802, 68629.3291332604, 49933.2126408239, 64358.5131164748, 102975.634845413, 84658.5787221384, 52243.061086781, 56988.7184089767, 43548.6558054764, 38536.145535317, 43115.4155553003, 50328.6208374212, 47191.4284861612, 42603.6038389972, 43749.2977769786, 43583.8022320709, 36672.3782472362, 26377.9033658583, 41746.6472558272, 26086.7868276075, 47228.5390765096, 75765.4757503756, 70939.107521347, 63949.8857556708, 55153.6701237129),timeStep=90)

Then I wrote a model with JAGS. The problem is that the CIR process has scaled noncentral chi square distribution and this is not directly supported in JAGS. So inspired by R implementation, I came up with 

model {
    for (i in 2:length(timeSerie)){
        ncp[i] <- 2*c*et*timeSerie[i-1]
        pd[i] ~ dpois(ncp[i]/2)
        cd[i] ~ dgamma(pd[i]+1E-10,c)
        gd[i] ~ dgamma(df/2,4*c)
        timeSerie[i] ~ dsum(cd[i],gd[i])
    }

    #parameters
    et <- exp(-kappa*timeStep)
    c <- 2*kappa/(sigma^2*(1-et))
    df <- 4*kappa*mu/sigma^2

    #CIR parameter priors 
    lkappa ~ dunif(-10,-2)
    kappa <- exp(lkappa)
    mu ~ dunif(30000,100000)
    lsigma ~ dunif(0,5) 
    sigma <- exp(lsigma)
}

Trying to run the model (jags.model in R), I get the following error 

  Error in node gd[18]
Unobserved node inconsistent with unobserved parents at initialization

(Now it is 18 instead of 8, because I use other simulated data).

I tried to debug JAGS (gdb and valgrind) to no avail: I don't have enough knowledge to do it…

Best Answer

I have received these errors when not using appropriate initial values for my parameters. Rather than writing functions to generate initial values (which I presume you have done), I would provide starting values that you know are consistent with one another.

Related Question