基于R语言模拟适配指定年份表的多元时间序列数据

Got it, let's work through how to simulate multivariate time series data that's perfectly aligned with your 1982-2008 year range. First, a quick catch: your current code generates 100 observations, but 1982 to 2008 is 27 years total—so we'll adjust the length first to match the time frame. Here's a step-by-step approach:

First, lock in the exact years you need and calculate how many observations that is:

# Set your target year range
time_range <- 1982:2008
n_years <- length(time_range) # This equals 27, which is our required sample size

For multivariate time series, you'll want variables that might correlate with each other (not just independent random numbers). The MASS::mvrnorm() function is perfect for this—it generates data from a multivariate normal distribution with a specified mean vector and covariance matrix.

Let's say you want 3 interrelated variables. Here's how to set it up:

# Load the MASS package (if you haven't already)
if (!require(MASS)) install.packages("MASS")

# Define means for each variable
mean_vec <- c(2, 5, 10) # e.g., Var1 has mean 2, Var2 mean 5, Var3 mean 10

# Define a covariance matrix to set correlations between variables
# Diagonals are variances, off-diagonals are covariances (which drive correlation)
cov_mat <- matrix(
  c(
    1.5^2, 0.6*1.5*2, 0.3*1.5*1, # Var1 variance + covariances with Var2/Var3
    0.6*1.5*2, 2^2, 0.4*2*1,     # Var2 variance + covariance with Var3
    0.3*1.5*1, 0.4*2*1, 1^2      # Var3 variance
  ),
  nrow = 3, ncol = 3
)

# Generate the multivariate data
multivariate_data <- MASS::mvrnorm(n = n_years, mu = mean_vec, Sigma = cov_mat)
colnames(multivariate_data) <- c("Var1", "Var2", "Var3")

Now, tie the simulated data directly to your 1982-2008 years. You can use either a standard data frame (great for general analysis) or a ts time series object (ideal for time-specific functions):

Option 1: Data Frame with Explicit Year Column #

df_data <- data.frame(Year = time_range, multivariate_data)
head(df_data) # Check the first few rows to confirm alignment

Option 2: Time Series (ts) Object #

ts_data <- ts(
  multivariate_data,
  start = min(time_range),
  end = max(time_range),
  frequency = 1 # Frequency = 1 because we're working with annual data
)
plot(ts_data) # Quick plot to visualize all variables over time

If you want to build off your original Z variable (adjusted to 27 observations) and add related variables, here's a simpler approach:

# Adjust your original Z to match the 27-year length
Z <- rnorm(n_years, mean = 0, sd = 1.5)

# Simulate Var2 with a strong positive correlation to Z
Var2 <- 0.7*Z + rnorm(n_years, mean = 3, sd = 1)

# Simulate Var3 with a weak negative correlation to Z
Var3 <- -0.2*Z + rnorm(n_years, mean = -1, sd = 2)

# Combine into a data frame
df_custom <- data.frame(Year = time_range, Z = Z, Var2 = Var2, Var3 = Var3)

Key Notes to Remember #

• Always double-check that your simulated data length matches the number of years in your range (27 for 1982-2008).
• Using a covariance matrix ensures your multivariate data has realistic relationships—skip this only if you want completely independent variables.
• ts objects make time-series-specific operations (like ARIMA modeling, trend analysis) much easier later on.

内容的提问来源于stack exchange，提问作者Stackuser