AgentSkillsCN

forecast

运用R语言中的forecast::auto.arima，对时间序列数据进行预测。当用户希望对数值型时间序列进行预测、预估未来值，或拟合ARIMA模型时，此技能将大显身手。但请勿将其用于分类任务、非时间序列回归，或当用户明确要求采用其他预测方法时。

SKILL.md
--- frontmatter
name: forecast
description: >
  Forecast time series data using R's forecast::auto.arima. Use when the user
  wants to forecast numeric time series, predict future values, or fit ARIMA models.
  Do NOT use for classification, non-time-series regression, or when the user
  explicitly requests a different forecasting method.
allowed-tools: Bash(Rscript *)

Forecast time series using R's forecast::auto.arima.

Quick Forecast

r
Rscript -e '
library(forecast)

# Read data (CSV with date and value columns)
data <- read.csv("INPUT_FILE")
ts_data <- ts(data$VALUE_COLUMN, frequency=FREQUENCY)

# Fit auto.arima with explicit settings and forecast
fit <- auto.arima(ts_data, ic="aic", stepwise=FALSE, lambda="auto")
fc <- forecast(fit, h=HORIZON)

# Print summary and forecasts
summary(fit)
print(fc)

# Save plot
png("forecast.png", width=800, height=400)
plot(fc, main="Forecast")
dev.off()

# Save forecasts to CSV
write.csv(data.frame(
  point=as.numeric(fc$mean),
  lo80=as.numeric(fc$lower[,1]),
  hi80=as.numeric(fc$upper[,1]),
  lo95=as.numeric(fc$lower[,2]),
  hi95=as.numeric(fc$upper[,2])
), "forecast_output.csv", row.names=FALSE)
'

Parameters

  • •FREQUENCY: Observations per cycle (12=monthly, 4=quarterly, 52=weekly, 365=daily)
  • •HORIZON: Number of periods to forecast ahead
  • •INPUT_FILE: Path to CSV with time series data
  • •VALUE_COLUMN: Name of the numeric column to forecast

Simulation

Use simulation for decisions under uncertainty — when you need the full distribution, not just point forecasts:

r
# Simulate 10k future paths from the fit above
sims <- simulate(fit, nsim=10000, future=TRUE)

# Decision-relevant summaries
cat("P(value > threshold):", mean(sims > THRESHOLD), "\n")
cat("Expected value:", mean(sims), "\n")
cat("5th/95th percentile:", quantile(sims, c(0.05, 0.95)), "\n")

Install R Dependencies

bash
Rscript -e 'install.packages("forecast", repos="https://cloud.r-project.org")'