nls（）错误-收敛失败：奇异收敛（7）

Question

我正在尝试为R中的论文评估异构资产定价模型。但是，当我运行代码时，我得到了错误；

Error in nls(P ~ PfitHV(P, Y, r, g, Beta, Phi1, Phi2, Delta, W), start = start.val,  : 
  Convergence failure: singular convergence (7)

我认为这可能与错误的起始值或某种类型的模型的过度参数化有关（基于关于同一错误的堆栈问题），但我不知道如何解决。我整日忙于此，所以任何帮助将不胜感激！

这是我的一些代码（很抱歉，但是我真的不知道要包括/排除什么）：

   memory <- T              # Use memory in fractions
    async.updating <- T     # Use asyncronous updating (and estimate delta)
    with.beta <- T          # Estimate beta (or fix it)

    # start values, lowerbound and upperbounds for phi1, delta_phi 
    start.val <- list(Phi1 = 0.2, Phi2 = 0.3)
    lower.val <- list(Phi1 = 0.0, Phi2 = 0.0)
    upper.val <- list(Phi1 = 1.5, Phi2 = 3.4)

    Beta<-1
    if (with.beta) {
      start.val <- c(start.val, Beta = 1)
      lower.val <- c(lower.val, Beta = 0)
      upper.val <- c(upper.val, Beta = 1000)
    }

    Delta <- 0.0
    if (async.updating) {
      start.val <- c(start.val, Delta = 0.5)
      lower.val <- c(lower.val, Delta = 0.0)
      upper.val <- c(upper.val, Delta = 1.0)
    }

    W <- 0
    if (memory) {
      start.val <- c(start.val, W = 0.1)
      lower.val <- c(lower.val, W = 0.0)
      upper.val <- c(upper.val, W = 1.0)
    }

    ##########################################################################################
    PfitHV <- function(P, Y, r, g, Beta, Phi1, Phi2, Delta, W) {
      T=NROW(P)  #number of rows (observations) 
      L=NCOL(P)  #number of columns

      # creating lags of P, Y
      P1 <- lag(P,k = 1)
      P2 <- lag(P,k = 2)
      Y1 <- lag(Y,k = 1)
      Y2 <- lag(Y,k = 2)

      # creation of Psi (in this version use static Gordon model)
    Psi <- (1+g)/(r-g)

      # payoffs of period t-1 -> equation 16 (without fraction in front)
      Pi1 <- (P1+Y1-(1+r)*P2)*((Phi1*Psi+1)*Phi1*(1+g)*Y2-(1+r)*P2)
      Pi2 <-(P1+Y1-(1+r)*P2)*((Phi2*Psi+1)*Phi2*(1+g)*Y2-(1+r)*P2)

     #memory in profit function 
      if (memory) {
        U1 <- filter((1-W)*Pi1, W, method="recursive")
        U2 <- filter((1-W)*Pi2, W, method="recursive")
      }
      else {
        U1 <- Pi1;
        U2 <- Pi2;
      }

      dUi <- U1-U2

      # normalize payoff differences
     dUi <- dUi/(sd(dUi)+0.0000001)

      # fractions of agents with rule 1, without asyncronous updating (general result)
      N1 <- 1/(1+exp(-Beta*dUi))

      # asynchronous updating
      if (async.updating) {
        N1 <-filter((1-Delta)*N1, Delta, method="recursive") 
      }
      else {
        N1 <- N1;
      }

      #predicted values
        N2 <- 1- N1;
        Q <- Psi*N1*Phi1+Psi*(N2)*Phi2

       #return value of Pfit (this is pricing equation)     
     Y*Q
    }

    ##########################################################################################
  estimate.model <- function(P, Y, r, g){
      library("nls2") 
      library('proto')
    # Use nonlinear least squares to estimate model 

      nlmod =
        nls(P~PfitHV(P, Y, r, g, Beta, Phi1, Phi2, Delta, W),
        start  = start.val,
        lower  = lower.val,
        upper  = upper.val,
        algorithm = "port",
        trace  = FALSE,
        control= nls.control(minFactor=1/10000, maxiter = 100))

      #retreive estimated parameter values
      d = summary(nlmod)
      Phi1 = d$coefficients[1];
      Phi2 = d$coefficients[2];
      Beta = d$coefficients[3];
      Delta = d$coefficients[4];
      W = d$coefficients[5];

      # redistil outcome variables using the fitted values 
      Pfit=PfitHV(P, Y, r, g, Beta, Phi1, Phi2, Delta, W); 
      res = P - Pfit;

      nlmod # outcome of function
    }

    #################main program#############################################################
    # read in data
    data <-  read.csv2('TRY2.csv', header=T, sep = ';', dec = ',', colClasses = "numeric", stringsAsFactors=FALSE)
    newdata <- na.omit(data) #to omit NA from data
    P <- newdata[,1] #imports prices 
    Y <- newdata[,2] #imports dividends
    G <- newdata[,18] #imports growth rate real cash flows
    Yield <- newdata[,17] #imports cash flow yield using dividends

    # calculate r,g averages
    Average_Yield <- mean(Yield)
    g <- mean(G)/100
    r <-g+Average_Yield/100

    nlmod=0; #clear all
    nlmod <- estimate.model(P, Y, r, g)
    summary(nlmod)

如果需要，我也可以共享我的数据文件。