Predicting Gold Prices: Backtesting of ML Models

Fitch projects a decline of about 30% in gold in 2026. Easing the trade war and the Israel-Iran conflict may support this idea. We will project how the prices could go by the end of the year.

We will use the modeltime.resample package for forecasting modeling.

library(tidymodels)
library(modeltime)
library(modeltime.resample)
library(tidyverse)
library(tidyquant)
library(timetk)

#Gold Futures
df_gold <- 
  tq_get("GC=F") %>% 
  select(date, close) %>% 
  drop_na()


#Make a Cross-Validation Training Plan
resamples_tscv <- 
  time_series_cv(
  data        = df_gold,
  assess      = "6 months",
  initial     = "5 years",
  skip        = "1 years",
  slice_limit = 4
)

#Begin with a Cross Validation Strategy
resamples_tscv %>%
  tk_time_series_cv_plan() %>%
  plot_time_series_cv_plan(date,
                           close, 
                           .facet_ncol = 2, 
                           .interactive = FALSE)


#Model 1: auto_arima
model_arima <- 
  arima_reg() %>%
  set_engine(engine = "auto_arima") %>% 
  fit(close ~ date, data = df_gold)

#Model 2: prophet
model_prophet <- 
  prophet_reg() %>%
  set_engine(engine = "prophet") %>% 
  fit(close ~ date, data = df_gold)

#Model 3: glmnet
model_glmnet <- 
  linear_reg(penalty = 0.2) %>%
  set_engine("glmnet")

rec_glmnet <- 
  recipe(close ~ ., data = df_gold) %>% 
  step_mutate(date_num = as.numeric(date)) %>% 
  step_date(date, features = "month") %>% 
  step_rm(date) %>% 
  step_dummy(all_nominal_predictors(), one_hot = TRUE) %>% 
  step_normalize(all_numeric_predictors()) 

glmnet_fit <- 
  workflow() %>% 
  add_recipe(rec_glmnet) %>% 
  add_model(model_glmnet) %>% 
  fit(df_gold)
  
#Modeltime Table
gold_models <- 
  modeltime_table(
    model_arima,
    model_prophet,
    glmnet_fit
  )

#Generate Resample Predictions
resamples_fitted <- 
  gold_models %>%
  modeltime_fit_resamples(
    resamples = resamples_tscv,
    control   = control_resamples(verbose = FALSE)
  )

#Accuracy Table
resamples_fitted %>%
  modeltime_resample_accuracy(summary_fns = mean) %>%
  table_modeltime_accuracy(.interactive = FALSE)


#Calibration for the Prophet Model 
calibration_prophet <- 
  model_prophet %>% 
  modeltime_calibrate(new_data = df_gold)

#Accuracy of the finalized model
calibration_prophet %>%
  modeltime_accuracy(metric_set = metric_set(rmse, rsq, mape))


#Forecast Forward
calibration_prophet %>% 
  modeltime_forecast(h = "6 months", 
                     actual_data = df_gold %>% 
                                   filter(date>= as.Date("2025-01-01"))) %>%
  plot_modeltime_forecast(.interactive = FALSE,
                          .legend_show = FALSE,
                          .line_size = 1.5,
                          .color_lab = "",
                          .title = "Gold Futures") +
  
  labs(subtitle = "<span style = 'color:dimgrey;'>Predictive Intervals</span><br><span style = 'color:red;'>ML Model</span>") + 
  scale_x_date(expand = expansion(mult = c(.1, .1)),
               labels = scales::label_date(format = "%b'%y")) +
  scale_y_continuous(labels = scales::label_currency()) +
  theme_minimal(base_family = "Roboto Slab", base_size = 20) +
  theme(legend.position = "none",
        plot.background = element_rect(fill = "azure", 
                                       color = "azure"),
        plot.title = element_text(face = "bold"),
        axis.text = element_text(face = "bold"),
        plot.subtitle = ggtext::element_markdown(face = "bold"))

< svg xmlns="http://www.w3.org/2000/svg" width="12" height="12" fill="none" viewBox="0 0 12 12"> < path fill="#fff" d="M2 0a2 2 0 0 0-2 2v2h1.5V2a.5.5 0 0 1 .5-.5h2V0H2Zm2 10.5H2a.5.5 0 0 1-.5-.5V8H0v2a2 2 0 0 0 2 2h2v-1.5ZM8 12v-1.5h2a.5.5 0 0 0 .5-.5V8H12v2a2 2 0 0 1-2 2H8Zm2-12a2 2 0 0 1 2 2v2h-1.5V2a.5.5 0 0 0-.5-.5H8V0h2Z" />

According to the Prophet model, the gold price seems to be difficult to reach the peak again by the end of the year.