Forecasting day-ahead electricity prices using a linear time series regression

Question

I have a dataset of hourly day-ahead electricity prices and hourly forecasted day-ahead demand (from governmental agency) in the Norwegian price area NO2. I am trying to use the forecasted day-ahead demand to predict the next 24-hour electricity prices for each day (cross validation using a rolling window) with a linear model. However, I am having trouble implementing the exogenous predictor (day-ahead demand) in this model without getting an error.

I have tried lagging the predictor, which looks like this:

# rolling window
NO2 <- NO2 %>% 
  filter(datetime > as.POSIXct("2023-01-01 00:00:00"))
NO2_stretch <- NO2 %>% 
  stretch_tsibble(.init = 48, .step = 24) %>% # initial training set of size 48 (two days), step = 24 as the training set should expand one day for each forecast iteration
  filter(.id !=max(.id)) # remove the last one as there will be no more observations to test
NO2_stretch # te "id" column corresponds to which row in the cross validation that we are in

# Estimate the linear models for each window
fit_cv <- NO2_stretch %>% 
  model(TSLM(no2_price ~ lag(no2_load, 24) + season())) # get a forecast for each day with an increasing training set

# can pipe this rolling window into forecast in the usual way to produce one step ahead forecasts from all models
fc_cv <- fit_cv %>% 
  forecast(h="1 day")

# Cross-validated accuracy
fc_cv %>%  accuracy(NO2_actual)

The forecast function gives me this error:

Error: Problem with `mutate()` input `TSLM(no2_price ~ no2_load + season())`.
x object 'no2_load' not found
  Unable to compute required variables from provided `new_data`.
  Does your model require extra variables to produce forecasts?
ℹ Input `TSLM(no2_price ~ no2_load + season())` is `(function (object, ...) ...`.

Any help on this would be highly appreciated, let me know if I can clarify this further.

Answer 1

At the moment you will need to provide a dataset containing the future values of no2_load to the forecast() function.

For example:

NO2_stretch_future <- NO2_stretch %>%
  new_data(n = 24) %>%
  mutate(no2_load = NA_real_)
fc_cv <- fit_cv %>% 
  forecast(new_data = NO2_stretch_future)

I haven't been able to test this code since you haven't provided a minimally reproducible example (MRE). Providing a MRE makes is easier for me to quickly and accurately answer your question.