GetECBHistoricalExchangeRateData <- function() {
    # 1. Setup temporary files and download the data
    temp_zip <- tempfile(fileext = ".zip")
    url <- "https://www.ecb.europa.eu/stats/eurofxref/eurofxref-hist.zip"

    message("Downloading data from ECB...")
    download.file(url, temp_zip, quiet = TRUE)

    # 2. Unzip and identify the CSV file
    unzipped_files <- unzip(temp_zip, exdir = tempdir())
    csv_file <- unzipped_files[grep("\\.csv$", unzipped_files)]

    # 3. Load the data into R
    message("Loading data...")
    df <- read.csv(csv_file, na.strings = c("N/A", "NA", ""), stringsAsFactors = FALSE)
    
    # Clean up temp files
    unlink(temp_zip)
    unlink(csv_file)
    
    return(df)
}
PreProcessData <- function(df, target_currencies, last_n_days = NULL) {
    
    # Preprocess Dates and Time
    df$Date <- as.Date(df$Date)
    df <- df[order(df$Date), ] 
	
	# Filter by the last N days if specified
	if (!is.null(last_n_days)) {
	  cutoff_date <- max(df$Date, na.rm = TRUE) - last_n_days
	  df <- df[df$Date >= cutoff_date, ]
	  message(sprintf("Filtered data to the last %d days (since %s).", last_n_days, cutoff_date))
	}

    # Create a Time Index for the regression
    df$TimeIndex <- as.numeric(df$Date - min(df$Date))
    
    # Ensure targets actually exist in the dataframe to prevent errors
    valid_targets <- intersect(target_currencies, names(df))

    # Filter the data frame down to Date, TimeIndex, and those specific targets
    df <- df[, c("Date", "TimeIndex", valid_targets)]

    # Create the new cross pairs dynamically
    # combn(valid_targets, 2) creates all unique 2-pair combinations
    if (length(valid_targets) >= 2) {
        pair_combos <- combn(valid_targets, 2)
        
        for (i in 1:ncol(pair_combos)) {
            base_curr <- pair_combos[1, i]
            quote_curr <- pair_combos[2, i]
            
            # Create column name, e.g., "USD_JPY"
            pair_name <- paste0(base_curr, "_", quote_curr)
            
            # Math: To get Base/Quote, divide the ECB Quote column by the ECB Base column
            df[[pair_name]] <- df[[quote_curr]] / df[[base_curr]]
        }
    }

    # Update the 'currencies' vector so the loop analyzes original and new columns
    currencies <- setdiff(names(df), c("Date", "TimeIndex"))
    
    return(list(currencies = currencies, df = df))
}
CalculateMetrics <- function(df,currencies) {
    # 5. Loop through each currency, run regression, and calculate metrics
    results_list <- list()

    message("Running regressions...")
    for (curr in currencies) {
      valid_data <- df[!is.na(df[[curr]]), ]
  
      if (nrow(valid_data) > 2) {
    
        # Simple linear regression: Exchange Rate ~ Time
        model <- lm(valid_data[[curr]] ~ valid_data$TimeIndex)
    
        # Extract Beta (slope) and Residual Standard Error (RSE)
        beta <- coef(model)[2]
        rse <- summary(model)$sigma 
    
        # Calculate Signal-to-Noise Ratio: |Beta| / RSE
        if (!is.na(rse) && rse > 0) {
          signal_to_noise <- abs(beta) / rse
        } else {
          signal_to_noise <- NA
        }
    
        # Store results
        results_list[[curr]] <- data.frame(
          Currency = curr,
          Beta = beta,
          AbsBeta = abs(beta),
          RSE = rse,
          SignalToNoise = signal_to_noise,
          stringsAsFactors = FALSE,
          row.names = NULL
        )
      }
    }

    # Combine all results
    results_df <- do.call(rbind, results_list)
    
    return (results_df)
    
}
OrderBySignalToNoiseRatio <- function(results_df) {
    
    # 6. Order by the Signal-to-Noise ratio and get the top 3
    top_3_currencies <- results_df[order(results_df$SignalToNoise, decreasing = TRUE), ]
    top_3_currencies <- head(top_3_currencies, 3)
    return (top_3_currencies)
}

#
#  10. RUNTIME
#

df               <- GetECBHistoricalExchangeRateData()
res              <- PreProcessData(df,c("USD", "JPY", "GBP","CAD","NOK"),720)
results_df       <- CalculateMetrics(res$df,res$currencies)
top_3_currencies <- OrderBySignalToNoiseRatio(results_df)

# Print the final output
message("\nTop 3 currencies by Trend Predictability (|Beta| / RSE):")
print(top_3_currencies)