roptimizationsolverportfoliocvxr

Issues with convex optimization and covariance matrix


I am using R studio and the CVXR package to optimize a portfolio. I have a separate function that optimizes the portfolio with the requested characteristics. The code has worked perfectly, but just now I tried to perform the code with inputs that have been calculated slightly differently.

For some reason, I now get the following error message depending on the inputs:

Error in construct_intermediate_chain(object, candidate_solvers, gp = gp) : Problem does not follow DCP rules.

The error message happens after running the solve-line of the code.As far as I understand, the error is related to the problem not fulfilling all the rules of convex optimization, but I don't understand the reason for this, as the same code has worked fine before.

The inputs are mu (expected returns vector), covareiance matrix, and a vector of previous weights for transaction costs. Here I have an example of inputs which did not work.

Covariance matrix:

                             Energy   Materials  Industrials Consumer Discretionary Consumer Staples   Health Care  Financials            IT Communication Services     Utilities  Real Estate
Energy                 0.0030897173 0.002379169 0.0023784794           0.0018105179     9.912975e-04  9.773525e-04 0.002644624  1.591055e-03           1.084336e-03  2.924996e-03 0.0019053206
Materials              0.0023791688 0.005259421 0.0033595945           0.0032327858     1.609355e-03  8.728170e-04 0.003128297  2.843850e-03           1.446845e-03  1.548390e-03 0.0027542633
Industrials            0.0023784794 0.003359594 0.0038704462           0.0030296529     9.267397e-04  1.174425e-03 0.003014640  4.269281e-03           2.002612e-03  2.007975e-03 0.0020530142
Consumer Discretionary 0.0018105179 0.003232786 0.0030296529           0.0043492339     4.066661e-04  6.214639e-04 0.002876835  5.505643e-03           2.923230e-03  9.575217e-04 0.0023722226
Consumer Staples       0.0009912975 0.001609355 0.0009267397           0.0004066661     2.166096e-03  1.377174e-03 0.001026166 -1.284662e-03           1.229443e-05  1.068868e-03 0.0011005523
Health Care            0.0009773525 0.000872817 0.0011744248           0.0006214639     1.377174e-03  2.613830e-03 0.001074507 -7.264024e-05           1.030068e-03  1.529622e-03 0.0008058046
Financials             0.0026446238 0.003128297 0.0030146396           0.0028768352     1.026166e-03  1.074507e-03 0.004320200  2.867530e-03           2.007997e-03  2.831038e-03 0.0021558241
IT                     0.0015910549 0.002843850 0.0042692808           0.0055056428    -1.284662e-03 -7.264024e-05 0.002867530  1.549390e-02           4.075777e-03  2.739874e-04 0.0026158084
Communication Services 0.0010843358 0.001446845 0.0020026123           0.0029232304     1.229443e-05  1.030068e-03 0.002007997  4.075777e-03           5.511017e-03 -7.925487e-06 0.0023098958
Utilities              0.0029249959 0.001548390 0.0020079747           0.0009575217     1.068868e-03  1.529622e-03 0.002831038  2.739874e-04          -7.925487e-06  5.636343e-03 0.0020081775
Real Estate            0.0019053206 0.002754263 0.0020530142           0.0023722226     1.100552e-03  8.058046e-04 0.002155824  2.615808e-03           2.309896e-03  2.008177e-03 0.0019311267

mu (expected returns):

         Energy    Materials  Industrials Consumer.Discretionary Consumer.Staples   Health.Care    Financials Information.Technology Communication.Services    Utilities  Real.Estate
33 -0.003017939 -0.006933182 -0.009011212            -0.01260558      0.002149697 -0.0008313939 -0.0008236061            -0.03658994            -0.03758739 -0.006729061 -0.008675182

previous_ws:

0.15 0.05 0.05 0.05 0.15 0.15 0.15 0.05 0.05 0.15 0.00

Now, the solver works if I use for example the following covariance matrix, which does not really make sense to me:

                             Energy    Materials  Industrials Consumer Discretionary Consumer Staples  Health Care  Financials           IT Communication Services    Utilities  Real Estate
Energy                 0.0029857628 0.0022884237 0.0017807713           0.0013177959     7.975600e-04 0.0007892713 0.001820128 1.553586e-03           0.0011628653 0.0017787287 0.0006446211
Materials              0.0022884237 0.0045078345 0.0027955769           0.0025927008     1.188452e-03 0.0006518281 0.002672642 2.757089e-03           0.0017992508 0.0011660224 0.0015354476
Industrials            0.0017807713 0.0027955769 0.0028927100           0.0025590297     9.795925e-04 0.0009694254 0.002486007 3.526496e-03           0.0021108035 0.0012218375 0.0011190885
Consumer Discretionary 0.0013177959 0.0025927008 0.0025590297           0.0035662042     9.633868e-04 0.0009565825 0.002743525 4.504239e-03           0.0028407505 0.0006230830 0.0012069240
Consumer Staples       0.0007975600 0.0011884525 0.0009795925           0.0009633868     1.943238e-03 0.0012036624 0.001524322 6.529904e-05           0.0007112165 0.0007460728 0.0003728238
Health Care            0.0007892713 0.0006518281 0.0009694254           0.0009565825     1.203662e-03 0.0023788680 0.001521153 1.093596e-03           0.0014957157 0.0007825620 0.0003020803
Financials             0.0018201278 0.0026726415 0.0024860071           0.0027435247     1.524322e-03 0.0015211531 0.003961717 2.808888e-03           0.0023976164 0.0015009125 0.0012940412
IT                     0.0015535858 0.0027570891 0.0035264962           0.0045042387     6.529904e-05 0.0010935961 0.002808888 1.147153e-02           0.0047435038 0.0003170311 0.0016258135
Communication Services 0.0011628653 0.0017992508 0.0021108035           0.0028407505     7.112165e-04 0.0014957157 0.002397616 4.743504e-03           0.0062305636 0.0005553472 0.0009298955
Utilities              0.0017787287 0.0011660224 0.0012218375           0.0006230830     7.460728e-04 0.0007825620 0.001500912 3.170311e-04           0.0005553472 0.0033356235 0.0012661852
Real Estate            0.0006446211 0.0015354476 0.0011190885           0.0012069240     3.728238e-04 0.0003020803 0.001294041 1.625813e-03           0.0009298955 0.0012661852 0.0020295775

And here is the code:

optimize_period <- function(mu, covar_matrix.period, previous_ws) {
  n.returns <- 0
  
  for (cell in mu){
    if (!is.na(cell)) {
      n.returns <- n.returns + 1
    } 
    else {
    }
  }

  print(n.returns)
  low_limit <- (1/n.returns)/2 
  high_limit <- (1/n.returns)*1.5 
  
  if (n.returns == 0) {
    print("return")
    return(0)
  } 
  previous_ws <- as.numeric(previous_ws)
  previous_ws.edit <- previous_ws[!is.na(previous_ws)]
  mu.edit <- mu[!is.na(mu)] 
  
  
  w <- Variable(n.returns)
  risk <- quad_form(w, covar_matrix.period)
  N <- 20 #Vaihdetaan 30 myƶhemmin
  constraints <- list(w >= low_limit, w <= high_limit, sum(w) == 1)
  gammas <- 10^seq(-3, 4, length.out = N)
  
  tc <- sum(abs(w - previous_ws.edit)) * transaction_cost
  
  mu
  
  ret <- (t(mu.edit) %*% w)
  
  rets <- rep(N, 0)
  stds <- rep(N, 0)
  tcs <- rep(N, 0)
  sharpes <- rep(N, 0)
  w_data <- matrix(0, nrow = N, ncol = n.returns)
  
 
  for (i in seq_along(gammas)) {
    gamma <- gammas[i]
    
    objective <- ret - tc - gamma * risk
    prob <- Problem(Maximize(objective), constraints) 
    result <- solve(prob, solver = "ECOS")
    
    rets[i] <- result$getValue(ret)
    stds[i] <- result$getValue(sqrt(risk))
    tcs[i] <- result$getValue(tc)
    w_data[i,] <- result$getValue(w)
    sharpes[i] <- rets[i]/stds[i]
    
  }
  
  max.row <- which.max(sharpes)
  opt_w <- (w_data[max.row,])
  
  for (cell in 1:length(mu)){
    if(is.na(mu[cell])) {
      opt_w[cell] <- 0
    }
  }
  print(opt_w)
  return(opt_w)
}

What is also confusing me is the fact that the crash happens when the optimization moves from 10 to 11 sectors to optimize, but this change has not had problems elsewhere. This is why I would guess the issue to be related to the Real Estate-column (as that was the new added sector). I tested this, and just changing the real estate column and real estate row fixes the issue, but I obviously have to use the real figures. My hunch is that for some reason the numbers in the covariance matrix happen to be unfortunate for the solver and make the problem seem non-convex, but I don't really have any clue as to why that would be. Any ideas how this could be worked around? Any help would be greatly appreciated!


Solution

  • Your covariance matrix is not positive definite. Note the negative eigenvalue. Here cc is the covariance matrix.

    eigen(cc)$values
    ##  [1]  0.0274352963  0.0121339336  0.0048068270  0.0037462092  0.0023325831
    ##  [6]  0.0012573482  0.0009386141  0.0008094488  0.0006894420  0.0005111973
    ## [11] -0.0004195685
    

    nearPD in Matrix will find the nearest positive definitive matrix although it still gives an eigenvalue that is close to zero so it may or may not be enough.

    library(Matrix)
    nearPD(cc)$eigenvalues
    ##  [1] 2.743530e-02 1.213393e-02 4.806827e-03 3.746209e-03 2.332583e-03
    ##  [6] 1.257348e-03 9.386141e-04 8.094488e-04 6.894420e-04 5.111973e-04
    ## [11] 2.743530e-10
    

    Note

    We provide cc in easily reproducible form

    cc <- 
    structure(c(0.0030897173, 0.0023791688, 0.0023784794, 0.0018105179, 
    0.0009912975, 0.0009773525, 0.0026446238, 0.0015910549, 0.0010843358, 
    0.0029249959, 0.0019053206, 0.002379169, 0.005259421, 0.003359594, 
    0.003232786, 0.001609355, 0.000872817, 0.003128297, 0.00284385, 
    0.001446845, 0.00154839, 0.002754263, 0.0023784794, 0.0033595945, 
    0.0038704462, 0.0030296529, 0.0009267397, 0.0011744248, 0.0030146396, 
    0.0042692808, 0.0020026123, 0.0020079747, 0.0020530142, 0.0018105179, 
    0.0032327858, 0.0030296529, 0.0043492339, 0.0004066661, 0.0006214639, 
    0.0028768352, 0.0055056428, 0.0029232304, 0.0009575217, 0.0023722226, 
    0.0009912975, 0.001609355, 0.0009267397, 0.0004066661, 0.002166096, 
    0.001377174, 0.001026166, -0.001284662, 1.229443e-05, 0.001068868, 
    0.001100552, 0.0009773525, 0.000872817, 0.001174425, 0.0006214639, 
    0.001377174, 0.00261383, 0.001074507, -7.264024e-05, 0.001030068, 
    0.001529622, 0.0008058046, 0.002644624, 0.003128297, 0.00301464, 
    0.002876835, 0.001026166, 0.001074507, 0.0043202, 0.00286753, 
    0.002007997, 0.002831038, 0.002155824, 0.001591055, 0.00284385, 
    0.004269281, 0.005505643, -0.001284662, -7.264024e-05, 0.00286753, 
    0.0154939, 0.004075777, 0.0002739874, 0.002615808, 0.001084336, 
    0.001446845, 0.002002612, 0.00292323, 1.229443e-05, 0.001030068, 
    0.002007997, 0.004075777, 0.005511017, -7.925487e-06, 0.002309896, 
    0.002924996, 0.00154839, 0.002007975, 0.0009575217, 0.001068868, 
    0.001529622, 0.002831038, 0.0002739874, -7.925487e-06, 0.005636343, 
    0.002008177, 0.0019053206, 0.0027542633, 0.0020530142, 0.0023722226, 
    0.0011005523, 0.0008058046, 0.0021558241, 0.0026158084, 0.0023098958, 
    0.0020081775, 0.0019311267), dim = c(11L, 11L), dimnames = list(
        c("Energy", "Materials", "Industrials", "Consumer_Discretionary", 
        "Consumer_Staples", "Health_Care", "Financials", "IT", "Communication_Services", 
        "Utilities", "Real_Estate"), c("Energy", "Materials", "Industrials", 
        "Consumer_Discretionary", "Consumer_Staples", "Health_Care", 
        "Financials", "IT", "Communication_Services", "Utilities", 
        "Real_Estate")))