Skip to content

Instantly share code, notes, and snippets.

@quantra-go-algo

quantra-go-algo/estimate_daily_arfima_forecasts.R Secret

Last active September 21, 2023 15:45
Download ZIP

Revisions

  1. quantra-go-algo revised this gist Sep 21, 2023. No changes.
  2. quantra-go-algo revised this gist Sep 21, 2023. 1 changed file with 73 additions and 14 deletions.
    87 changes: 73 additions & 14 deletions estimate_daily_arfima_forecasts.R
    View file
    Original file line number Diff line number Diff line change
    @@ -1,29 +1,88 @@
    # Set the number of cores to be used for parallelize the ARFIMA estimation
    nCores <- detectCores() - 1

    # Make the cluster based on the previous number of cores
    cluster <- makePSOCKcluster(nCores)
    # A function to estimate the ARFIMA model
    arfima_func <- function(data) {

    # Export the arfima_fit cluster to be used in the for loop
    clusterExport(cluster, 'arfima_fit')
    # Estimate the ARFIMA model
    arfima_fit <- autoarfima(data, ar.max = 5, ma.max = 5, criterion = "BIC", method = "partial",
    arfima = TRUE, include.mean = TRUE, cluster = cluster,
    distribution.model = "norm", solver = "hybrid", return.all = FALSE)
    return(arfima_fit)
    }

    # A function to stop runningn the ARFIMA model in case it takes more 10 minutes to run
    myWrapperFunction <- function(data) {
    result = tryCatch(expr = withTimeout(arfima_func(data), timeout = 10*60),
    TimeoutException = function(ex) "TimedOut")
    return(result)
    }

    # The for loop to estimate the ARFIMA model each day
    for (i in initial_iloc_to_forecast:nrow(data)) {
    for (i in (initial_iloc_to_forecast):nrow(data)) {
    start_time <- Sys.time()

    # Select the corresponding 6-year previous data to train the ARFIMA model
    #data_sample <- subset(data, !(data$date %in% dates_to_be_removed))
    data_sample <- data$AAPL[((i-1)-(span-1)):(i-1)]

    # Estimate the ARFIMA model
    arfima_fit <- autoarfima(data_sample, ar.max = 5, ma.max = 5, criterion = "BIC", method = "partial",
    arfima = TRUE, include.mean = TRUE, cluster = cluster,
    distribution.model = "norm", solver = "hybrid", return.all = FALSE)
    # Print some days info
    print(paste0(strrep('=',50)))
    print(paste0("Forecast ",(i+1-initial_iloc_to_forecast), ' out of ', length(df_forecasts$date)))

    # Make the cluster based on the previous number of cores
    cluster <- makePSOCKcluster(nCores)

    # Estimate the ARFIMA model for this day
    arfima_fit <- myWrapperFunction(data_sample)

    # Forecast the next day Apple stock close price
    forecast <- arfimaforecast(arfima_fit$fit, data=data_sample, n.ahead=1)@forecast$seriesFor[1]
    # Stop the CPU multithreading
    stopCluster(cluster)

    # Go long if the forecast is higher than the previous-day stock close price
    df_forecasts[(i+1-initial_iloc_to_forecast),'signal'] <-
    if (forecast>=data_sample[length(data_sample)]) 1 else 0
    if (typeof(arfima_fit)!="list") {

    # Print a statement saying estimtation took too long
    print(paste0("Estimation took longer than expected. Today's signal will be 0."))

    # Signal will be 0 this day
    df_forecasts[(i+1-initial_iloc_to_forecast),'signal'] <- 0.0

    # Save the df_forecasts in an excel file
    write.xlsx(df_forecasts, 'df_forecasts_in_R_v01.xlsx')

    # Print some important things for log
    print(paste0("Date is ",df_forecasts[(i+1-initial_iloc_to_forecast),]$date))
    print(paste0("Signal is ",df_forecasts[(i+1-initial_iloc_to_forecast),'signal']))
    print(paste0("Start time is ",start_time))
    end_time <- Sys.time()
    print(paste0("End time is ",end_time))

    #calculate time difference in minutes
    diff <- difftime(end_time, start_time, units="mins")
    print(paste0("Time passed is ",round(diff,2)," minutes"))
    } else {
    # Obtain the data sample Apple last price
    last_apple_price <- tail(data_sample,1)[[1]]

    # Forecast the next day Apple stock close price forecast
    forecast <- arfimaforecast(arfima_fit$fit, data=data_sample, n.ahead=1)@forecast$seriesFor[1]

    # Go long if the forecast is higher than the previous-day stock close price
    df_forecasts[(i+1-initial_iloc_to_forecast),'signal'] <- if (forecast>=last_apple_price) 1 else 0

    # Save the df_forecasts in an excel file
    write.xlsx(df_forecasts, 'df_forecasts_in_R_v01.xlsx')

    print(paste0("Date is ",df_forecasts[(i+1-initial_iloc_to_forecast),]$date))
    print(paste0('Price forecast is ',forecast))
    print(paste0('Last AAPL price is ',last_apple_price))
    print(paste0("Signal is ",df_forecasts[(i+1-initial_iloc_to_forecast),'signal']))
    print(paste0("Start time is ",start_time))
    end_time <- Sys.time()
    print(paste0("End time is ",end_time))

    #calculate time difference in minutes
    diff <- difftime(end_time, start_time, units="mins")
    print(paste0("Time passed is ",round(diff,2)," minutes"))
    }
    }
  3. quantra-go-algo created this gist Aug 11, 2023.
    29 changes: 29 additions & 0 deletions estimate_daily_arfima_forecasts.R
    View file
    Original file line number Diff line number Diff line change
    @@ -0,0 +1,29 @@
    # Set the number of cores to be used for parallelize the ARFIMA estimation
    nCores <- detectCores() - 1

    # Make the cluster based on the previous number of cores
    cluster <- makePSOCKcluster(nCores)

    # Export the arfima_fit cluster to be used in the for loop
    clusterExport(cluster, 'arfima_fit')

    # The for loop to estimate the ARFIMA model each day
    for (i in initial_iloc_to_forecast:nrow(data)) {
    start_time <- Sys.time()

    # Select the corresponding 6-year previous data to train the ARFIMA model
    data_sample <- data$AAPL[((i-1)-(span-1)):(i-1)]

    # Estimate the ARFIMA model
    arfima_fit <- autoarfima(data_sample, ar.max = 5, ma.max = 5, criterion = "BIC", method = "partial",
    arfima = TRUE, include.mean = TRUE, cluster = cluster,
    distribution.model = "norm", solver = "hybrid", return.all = FALSE)

    # Forecast the next day Apple stock close price
    forecast <- arfimaforecast(arfima_fit$fit, data=data_sample, n.ahead=1)@forecast$seriesFor[1]

    # Go long if the forecast is higher than the previous-day stock close price
    df_forecasts[(i+1-initial_iloc_to_forecast),'signal'] <-
    if (forecast>=data_sample[length(data_sample)]) 1 else 0

    }