New Simulator

We have updated annual simulator. The current version uses two factors: S&P 500 volatility and BAA bond rate. Initial factors are from May 2025. We simulate portfolio of three classes of assets:

  • Standard & Poor 500 USA Stocks
  • MSCI EAFE Developed Markets Index Stocks
  • USA Investment-Grade Corporate Bonds, ICE Bank of America Index

We use the following autoregression equations for factors:

 \ln V(t) = \alpha + \beta \ln V(t) + W(t) with \alpha = 0.847850 and \beta = 0.620146.

 \ln R(t) = \gamma + \theta \ln R(t-1) + Z(t) with  \gamma = 0.107208 and \theta = 0.942062.

For corporate bonds, US stocks, and international stocks, denote their annual arithmetic total returns in % as  Q_0, Q_1, Q_2

  •  Q_0(t) = a_0 - c_0(R(t) - R(t-1)) + \delta_0(t) with  a_0 = -1.104258 and  c_0 = 6.040179
  •  Q_1(t) = a_1 - b_1V(t) - c_1(R(t) - R(t-1)) + \delta_1(t) with  a_1 = 22.529302 and  b_1 = 1.004078 and  c_1 = 6.755580.
  •  Q_2(t) = a_2 - b_2V(t) - c_2(R(t) - R(t-1)) + \delta_2(t) with  a_2 = 29.408882 and  b_2 = 1.104258 and  c_2 = 4.905754.

All five series of residuals  \delta_0, \delta_1, \delta_2, Z, W are well-modeled by independent identically distributed random variables, judging by Monte Carlo simulation. Unfortunately, they are not normal. Namely,  Z and  W (the innovations for factor autoregressions) are closer to skew-normal. I did not yet pursue this direction of research. But the other three residual series  \delta_0, \delta_1, \delta_2 are Gaussian.

However, I still modeled these five series as multivariate Gaussian with mean vector zero and empirical covariance matrix

 \begin{bmatrix} 0.132753 & -0.065028 & -0.043977 & 0.014952 & 0.129096 \\ -0.065028 & 2.82715 & 0.98068 & -0.027136 & 1.84788 \\ -0.043977 & 0.98068 & 3.979577 & -0.000627 & -0.281541 \\ 0.014952 & -0.027136 & -0.000627 & 0.01842 & -0.030759 \\ 0.129096 & 1.84788 & -0.281541 & -0.030759 & 8.115026 \\ \end{bmatrix}

We consider only nominal, not real returns. To compensate for that, withdrawals/contributions can change annually.

We allow for constant split between US and international stocks. Bond and overall stock percentages might change from year to year linearly. This is to allow for a more conservative portfolio as time goes.

We use May 2025 averages for initial factors: volatility (using rescaled VIX) and rate.

See the https://github.com/asarantsev/simulator-current



Published by


Response

  1. Updated Simulator for Rate and Volatility – My Finance

    […] the previous post, we updated the financial simulator to make for log returns instead of arithmetic returns. We had […]

    Like

Leave a comment