Introduction

This analysis explores 100 years of global fuel pricing data (1924-2024), examining gasoline, diesel, and LPG prices across 25 countries and 9 regions. The dataset includes inflation-adjusted prices, crude oil benchmarks, tax rates, subsidy indicators, and macroeconomic variables. Rather than simply describing the data, each question builds on the last to uncover why prices differ, when they change, and what drives those changes.

Load Necessary Libraries and Dataset

library(readr)
library(dplyr)
library(tidyr)
library(ggplot2)
library(GGally)
library(corrplot)
library(scales)

fuel_prices <- read.csv("fuel_prices/all_countries_combined.csv")
glimpse(fuel_prices)
## Rows: 2,525
## Columns: 27
## $ year                     <int> 1924, 1925, 1926, 1927, 1928, 1929, 1930, 193…
## $ country                  <chr> "Argentina", "Argentina", "Argentina", "Argen…
## $ iso3                     <chr> "ARG", "ARG", "ARG", "ARG", "ARG", "ARG", "AR…
## $ currency_code            <chr> "ARS", "ARS", "ARS", "ARS", "ARS", "ARS", "AR…
## $ region                   <chr> "Latin America", "Latin America", "Latin Amer…
## $ decade                   <chr> "1920s", "1920s", "1920s", "1920s", "1920s", …
## $ gasoline_usd_per_liter   <dbl> 0.0124, 0.0121, 0.0125, 0.0126, 0.0131, 0.012…
## $ gasoline_usd_per_gallon  <dbl> 0.0469, 0.0458, 0.0473, 0.0477, 0.0496, 0.048…
## $ gasoline_real_2024usd    <dbl> 0.2277, 0.2171, 0.2218, 0.2274, 0.2405, 0.235…
## $ diesel_usd_per_liter     <dbl> 0.0101, 0.0098, 0.0102, 0.0103, 0.0108, 0.010…
## $ diesel_real_2024usd      <dbl> 0.1855, 0.1758, 0.1809, 0.1859, 0.1983, 0.189…
## $ lpg_usd_per_liter        <dbl> 0.0059, 0.0056, 0.0062, 0.0062, 0.0064, 0.006…
## $ gasoline_local_currency  <dbl> 11.16, 10.89, 11.25, 11.34, 11.79, 11.52, 11.…
## $ gasoline_yoy_pct_change  <dbl> NA, -2.42, 3.31, 0.80, 3.97, -2.29, 1.56, 5.3…
## $ diesel_yoy_pct_change    <dbl> NA, -2.97, 4.08, 0.98, 4.85, -4.63, 4.85, 0.0…
## $ price_tier               <chr> "Very Low (<$0.30)", "Very Low (<$0.30)", "Ve…
## $ tax_pct_of_pump_price    <dbl> 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 1…
## $ pretax_price_usd_ltr     <dbl> 0.0107, 0.0104, 0.0108, 0.0108, 0.0113, 0.011…
## $ subsidy_flag             <int> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ subsidy_regime           <chr> "heavy", "heavy", "heavy", "heavy", "heavy", …
## $ is_oil_producer          <int> 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, …
## $ crude_oil_usd_per_barrel <dbl> 1.43, 1.68, 1.88, 1.30, 1.17, 1.27, 1.19, 0.6…
## $ crude_oil_usd_per_liter  <dbl> 0.0090, 0.0106, 0.0118, 0.0082, 0.0074, 0.008…
## $ refinery_margin_usd_ltr  <dbl> 0.0034, 0.0015, 0.0007, 0.0044, 0.0057, 0.004…
## $ oil_price_shock_idx      <dbl> 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, …
## $ us_cpi                   <dbl> 17.1, 17.5, 17.7, 17.4, 17.1, 17.1, 16.7, 15.…
## $ inflation_deflator_2024  <dbl> 18.3626, 17.9429, 17.7401, 18.0460, 18.3626, …

Phase 1: Understanding the Data

Question 1.1: What is the structure and scope of the dataset?

Why this question? Before any analysis, we must confirm what data we have, its temporal and geographic coverage, and whether variable types are appropriate for the analysis we plan to do.

str(fuel_prices)
## 'data.frame':    2525 obs. of  27 variables:
##  $ year                    : int  1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 ...
##  $ country                 : chr  "Argentina" "Argentina" "Argentina" "Argentina" ...
##  $ iso3                    : chr  "ARG" "ARG" "ARG" "ARG" ...
##  $ currency_code           : chr  "ARS" "ARS" "ARS" "ARS" ...
##  $ region                  : chr  "Latin America" "Latin America" "Latin America" "Latin America" ...
##  $ decade                  : chr  "1920s" "1920s" "1920s" "1920s" ...
##  $ gasoline_usd_per_liter  : num  0.0124 0.0121 0.0125 0.0126 0.0131 0.0128 0.013 0.0137 0.0137 0.0145 ...
##  $ gasoline_usd_per_gallon : num  0.0469 0.0458 0.0473 0.0477 0.0496 0.0485 0.0492 0.0519 0.0519 0.0549 ...
##  $ gasoline_real_2024usd   : num  0.228 0.217 0.222 0.227 0.24 ...
##  $ diesel_usd_per_liter    : num  0.0101 0.0098 0.0102 0.0103 0.0108 0.0103 0.0108 0.0108 0.011 0.0116 ...
##  $ diesel_real_2024usd     : num  0.185 0.176 0.181 0.186 0.198 ...
##  $ lpg_usd_per_liter       : num  0.0059 0.0056 0.0062 0.0062 0.0064 0.0061 0.0061 0.0068 0.0066 0.007 ...
##  $ gasoline_local_currency : num  11.2 10.9 11.2 11.3 11.8 ...
##  $ gasoline_yoy_pct_change : num  NA -2.42 3.31 0.8 3.97 -2.29 1.56 5.38 0 5.84 ...
##  $ diesel_yoy_pct_change   : num  NA -2.97 4.08 0.98 4.85 -4.63 4.85 0 1.85 5.45 ...
##  $ price_tier              : chr  "Very Low (<$0.30)" "Very Low (<$0.30)" "Very Low (<$0.30)" "Very Low (<$0.30)" ...
##  $ tax_pct_of_pump_price   : num  14 14 14 14 14 14 14 14 14 14 ...
##  $ pretax_price_usd_ltr    : num  0.0107 0.0104 0.0108 0.0108 0.0113 0.011 0.0112 0.0118 0.0118 0.0125 ...
##  $ subsidy_flag            : int  0 0 0 0 0 0 0 0 0 0 ...
##  $ subsidy_regime          : chr  "heavy" "heavy" "heavy" "heavy" ...
##  $ is_oil_producer         : int  1 1 1 1 1 1 1 1 1 1 ...
##  $ crude_oil_usd_per_barrel: num  1.43 1.68 1.88 1.3 1.17 1.27 1.19 0.65 0.87 0.67 ...
##  $ crude_oil_usd_per_liter : num  0.009 0.0106 0.0118 0.0082 0.0074 0.008 0.0075 0.0041 0.0055 0.0042 ...
##  $ refinery_margin_usd_ltr : num  0.0034 0.0015 0.0007 0.0044 0.0057 0.0048 0.0055 0.0096 0.0082 0.0103 ...
##  $ oil_price_shock_idx     : num  1 1 1 1 1 1 1 1 1 1 ...
##  $ us_cpi                  : num  17.1 17.5 17.7 17.4 17.1 17.1 16.7 15.2 13.7 13 ...
##  $ inflation_deflator_2024 : num  18.4 17.9 17.7 18 18.4 ...
cat("Dataset Dimensions:", nrow(fuel_prices), "rows x",
    ncol(fuel_prices), "columns\n")
## Dataset Dimensions: 2525 rows x 27 columns
cat("Time Period:", min(fuel_prices$year),
    "to", max(fuel_prices$year), "\n")
## Time Period: 1924 to 2024
cat("Countries Covered:", n_distinct(fuel_prices$country), "\n")
## Countries Covered: 25
cat("Regions Covered:", n_distinct(fuel_prices$region), "\n")
## Regions Covered: 9

Key findings:

  • 2,525 observations spanning 100 years (1924-2024)
  • 27 variables including nominal prices, inflation-adjusted prices, crude oil benchmarks, tax rates, and policy indicators
  • 25 countries across 9 global regions
  • Mix of developed economies (Europe, North America) and emerging markets (Latin America, Asia, Middle East)
  • Data types are appropriate: numeric for prices, character for categorical groupings

Implications:

  • The century-long timeframe allows analysis of historical shocks (Great Depression, oil crises, 2008 financial crisis, COVID-19)
  • Multi-country coverage enables comparative policy analysis across tax and subsidy regimes

Question 1.2: Are there missing values that could affect our analysis?

Why this question? Missing data can bias results or limit certain analyses. Understanding gaps helps determine appropriate analytical methods.

missing_summary <- colSums(is.na(fuel_prices))
missing_df <- data.frame(
  Variable        = names(missing_summary),
  Missing_Count   = missing_summary,
  Percent_Missing = round((missing_summary / nrow(fuel_prices)) * 100, 2)
) %>%
  filter(Missing_Count > 0) %>%
  arrange(desc(Missing_Count))

print(missing_df)
##                                        Variable Missing_Count Percent_Missing
## gasoline_yoy_pct_change gasoline_yoy_pct_change            25            0.99
## diesel_yoy_pct_change     diesel_yoy_pct_change            25            0.99

Key findings:

  • Only 25 missing values in gasoline_yoy_pct_change and diesel_yoy_pct_change (0.99% of data)
  • These are the first year (1924) for each country where year-over-year comparison is impossible — structural missingness, not data quality issues
  • All other 25 variables have complete coverage

Implications:

  • The dataset is exceptionally clean with 99%+ completeness
  • No imputation needed
  • Year-over-year analyses will simply exclude 1924

Question 1.3: What is the central tendency and spread of gasoline prices — and is the mean misleading?

Why this question? Mean alone is not enough. When mean and median are very different, the data is skewed — meaning the average is being pulled by extreme values and does not represent a typical observation. Standard deviation tells us how spread out prices are across the full 100-year range.

fuel_prices %>%
  summarise(
    Mean_Nominal   = round(mean(gasoline_usd_per_liter), 4),
    Median_Nominal = round(median(gasoline_usd_per_liter), 4),
    SD_Nominal     = round(sd(gasoline_usd_per_liter), 4),
    Mean_Real      = round(mean(gasoline_real_2024usd), 4),
    Median_Real    = round(median(gasoline_real_2024usd), 4),
    SD_Real        = round(sd(gasoline_real_2024usd), 4),
    Min_Real       = round(min(gasoline_real_2024usd), 4),
    Max_Real       = round(max(gasoline_real_2024usd), 4)
  )
##   Mean_Nominal Median_Nominal SD_Nominal Mean_Real Median_Real SD_Real Min_Real
## 1       0.3001         0.0551     0.4582    0.6725      0.4528  0.5686   0.0123
##   Max_Real
## 1   2.6502

Key findings:

  • Nominal mean ($0.30) vs median ($0.055): The mean is nearly 6x the median — the distribution is heavily right-skewed. A small number of recent high-price observations in modern Europe are pulling the average up dramatically
  • Real mean ($0.67) vs median ($0.45): Even after inflation adjustment, mean still exceeds median — still right-skewed, but less extreme
  • Standard deviation of $0.56 (real): Very high relative to the mean of $0.67 — prices vary enormously across countries and time
  • Real prices are 2.2x higher than nominal, showing the true impact of century-long inflation

Implications:

  • The median ($0.45/L real) is the more honest “typical” global fuel price — the mean is distorted by outliers
  • All subsequent analyses use real prices for accurate historical comparisons
  • A gallon costing $0.20 in 1950 had similar purchasing power to $1.80+ today

Question 1.4: What is the mode of the price tier — which pricing bracket do most country-years fall into?

Why this question? Price tier is a categorical variable. The mode (most frequent category) tells us what “normal” looks like globally across the full dataset. This is especially useful when the numeric mean is misleading due to skewness.

tier_counts <- fuel_prices %>%
  count(price_tier, sort = TRUE) %>%
  mutate(Percentage = round(n / sum(n) * 100, 1))

print(tier_counts)
##             price_tier    n Percentage
## 1    Very Low (<$0.30) 1790       70.9
## 2    Low ($0.30–$0.70)  305       12.1
## 3 Medium ($0.70–$1.10)  207        8.2
## 4   High ($1.10–$1.60)  162        6.4
## 5   Very High (>$1.60)   61        2.4
cat("\nMode (most frequent price tier):", tier_counts$price_tier[1], "\n")
## 
## Mode (most frequent price tier): Very Low (<$0.30)
cat("Appearing in", tier_counts$n[1], "observations (",
    tier_counts$Percentage[1], "% of all data)\n")
## Appearing in 1790 observations ( 70.9 % of all data)

Key findings:

  • The modal price tier is “Very Low (<$0.30)”, the most common single bracket across all country-years
  • This reflects that most of the dataset is historical (pre-1970s) when prices were genuinely very low in real terms, plus modern subsidised markets
  • “High” and “Very High” tiers represent modern developed economies and are a minority of observations

Implications:

  • Globally and historically, cheap fuel is the norm — expensive fuel is a modern, policy-driven phenomenon
  • The right-skewed distribution and the mode together confirm the mean overstates the “typical” experience

Question 1.5: What is the historical range of crude oil prices?

Why this question? Crude oil is the primary input cost for gasoline and diesel. Understanding its volatility contextualises all retail price movements throughout the century.

oil_summary <- fuel_prices %>%
  summarise(
    Min_Oil    = min(crude_oil_usd_per_barrel),
    Max_Oil    = max(crude_oil_usd_per_barrel),
    Mean_Oil   = round(mean(crude_oil_usd_per_barrel), 2),
    Median_Oil = round(median(crude_oil_usd_per_barrel), 2),
    SD_Oil     = round(sd(crude_oil_usd_per_barrel), 2)
  )

print(oil_summary)
##   Min_Oil Max_Oil Mean_Oil Median_Oil SD_Oil
## 1    0.65  112.01       23       9.35  29.98
# Identify years of extreme crude prices
fuel_prices %>%
  filter(crude_oil_usd_per_barrel == min(crude_oil_usd_per_barrel) |
         crude_oil_usd_per_barrel == max(crude_oil_usd_per_barrel)) %>%
  select(year, country, crude_oil_usd_per_barrel) %>%
  distinct(year, .keep_all = TRUE)
##   year   country crude_oil_usd_per_barrel
## 1 1931 Argentina                     0.65
## 2 2012 Argentina                   112.01

Key findings:

  • Minimum: $0.65/barrel (1931, Great Depression era)
  • Maximum: $112.01/barrel (2012, post-2008 recovery peak)
  • Mean: $23.00/barrel vs Median: $9.35/barrel — again heavily right-skewed by recent high prices
  • Standard deviation ~$24 shows extreme historical price swings

Implications:

  • A 172x difference between lowest and highest crude prices over the century
  • Cheap oil era (pre-1970): median ~$2-3/barrel
  • Oil shock era (1973-1985): prices jumped 10x in a decade
  • Modern era (2000s onward): sustained prices above $50/barrel
  • The mean-median gap mirrors what we saw in retail prices — the same skewness pattern appears at every level of the supply chain

Question 1.6: How do the key economic variables correlate and distribute simultaneously?

Why this question? Before building predictive models, we need to understand how our numerical variables interact with one another. A pair plot visually displays the distribution shape of each variable while simultaneously mapping the scatterplots and correlation coefficients for every variable pairing, instantly revealing multi-collinearity and non-linear trends..

cor_vars <- fuel_prices %>%
  select(gasoline_real_2024usd, crude_oil_usd_per_barrel, 
         tax_pct_of_pump_price, us_cpi) %>%
  na.omit()

ggpairs(cor_vars, 
        title = "Pair Plot of Key Economic Indicators",
        lower = list(continuous = wrap("points", alpha = 0.3, size = 0.5, color = "steelblue")),
        diag = list(continuous = wrap("densityDiag", fill = "orange", alpha = 0.5))) +
  theme_minimal()

Key findings:

  • The diagonal density plots visually confirm the heavy right-skew in real gasoline prices and crude oil prices.
  • The scatterplots between crude_oil_usd_per_barrel and gasoline_real_2024usd show a positive relationship, but the wide spread indicates other factors (like taxes) heavily influence the final price.
  • The relationship between tax rates and real gasoline prices shows distinct clustering, pointing to different regional policy models.

Implications:

  • Because the variables are right-skewed and some relationships appear curved rather than strictly linear, this justifies the need for both linear and polynomial regression modeling later in the analysis to capture the true complexity of the market.

Phase 2: Comparative Analysis

Question 2.1: Do oil-producing countries actually have cheaper fuel than non-producers?

Why this question? This tests a logical hypothesis — countries that extract oil should face lower domestic prices due to direct supply access and subsidies. The data either confirms or challenges this assumption.

producer_stats <- fuel_prices %>%
  mutate(Producer = ifelse(is_oil_producer == 1,
                           "Oil Producer", "Non-Producer")) %>%
  group_by(Producer) %>%
  summarise(
    Mean_Real_Price   = round(mean(gasoline_real_2024usd), 3),
    Median_Real_Price = round(median(gasoline_real_2024usd), 3),
    SD_Real_Price     = round(sd(gasoline_real_2024usd), 3),
    Mean_Tax_Rate     = round(mean(tax_pct_of_pump_price), 1),
    Subsidy_Pct       = round(mean(subsidy_flag) * 100, 1),
    Count             = n()
  )

print(producer_stats)
## # A tibble: 2 × 7
##   Producer     Mean_Real_Price Median_Real_Price SD_Real_Price Mean_Tax_Rate
##   <chr>                  <dbl>             <dbl>         <dbl>         <dbl>
## 1 Non-Producer           1.07              0.856         0.601          43.7
## 2 Oil Producer           0.517             0.35          0.471          20.7
## # ℹ 2 more variables: Subsidy_Pct <dbl>, Count <int>
fuel_prices %>%
  mutate(Producer = ifelse(is_oil_producer == 1,
                           "Oil Producer", "Non-Producer")) %>%
  ggplot(aes(x = Producer, y = gasoline_real_2024usd, fill = Producer)) +
  geom_boxplot(outlier.alpha = 0.3, show.legend = FALSE) +
  labs(title = "Real Gasoline Price: Oil Producers vs Non-Producers",
       subtitle = "Inflation-adjusted to 2024 USD",
       x = "",
       y = "Real Price (USD/Liter)") +
  scale_fill_manual(values = c("Oil Producer"  = "darkgreen",
                                "Non-Producer" = "steelblue")) +
  theme_minimal()

Key findings:

  • Oil producers have significantly lower median prices (~$0.25-0.30/L) compared to non-producers (~$0.55-0.60/L)
  • Producers also have much lower average tax rates and higher subsidy usage
  • However, Norway (oil producer, high-tax) is as expensive as Germany (non-producer, high-tax)

Implications:

  • Being an oil producer alone does not guarantee cheap fuel — policy choice (subsidy vs. tax) is the decisive factor
  • Norway proves this clearly: a nation can produce oil and still maintain high consumer prices through taxation

Question 2.2: Which countries have had the highest fuel prices historically?

Why this question? Identifying consistently expensive markets reveals the long-run impact of taxation, geography, and energy policy — not just short-term price spikes.

top_expensive <- fuel_prices %>%
  group_by(country) %>%
  summarise(
    Avg_Real_Gasoline = round(mean(gasoline_real_2024usd), 3),
    Avg_Tax_Percent   = round(mean(tax_pct_of_pump_price), 1),
    Is_Oil_Producer   = max(is_oil_producer)
  ) %>%
  arrange(desc(Avg_Real_Gasoline)) %>%
  head(10)

print(top_expensive)
## # A tibble: 10 × 4
##    country        Avg_Real_Gasoline Avg_Tax_Percent Is_Oil_Producer
##    <chr>                      <dbl>           <dbl>           <int>
##  1 Norway                     1.42             47.2               1
##  2 Italy                      1.29             50.3               0
##  3 France                     1.24             49.5               0
##  4 Germany                    1.23             48                 0
##  5 Singapore                  1.09             36.5               0
##  6 United Kingdom             1.06             47.2               1
##  7 South Korea                0.978            39.6               0
##  8 Japan                      0.894            36.5               0
##  9 Turkey                     0.798            45.7               0
## 10 Australia                  0.763            28.9               1

Key findings:

  • Norway ($1.42/L): Highest average real price despite being an oil producer — environmental taxation policy dominates
  • Italy ($1.29/L), France ($1.24/L), Germany ($1.23/L): European high-tax model
  • Singapore ($1.09/L): Island nation with full import dependence
  • All top 10 have average tax rates of 40-60% of pump price

Implications:

  • High prices correlate with environmental policy, not just scarcity
  • Geography matters: island nations face higher import costs
  • Policy choice, not geology, determines long-term price levels

Question 2.3: Which region has the highest price volatility, and why does that matter?

Why this question? Volatility (measured by Coefficient of Variation) measures economic risk and instability. High volatility means consumers face unpredictable costs, which disrupts household budgets and business planning far more than a consistently high-but-stable price.

regional_volatility <- fuel_prices %>%
  group_by(region) %>%
  summarise(
    Mean_Price   = round(mean(gasoline_real_2024usd), 3),
    Median_Price = round(median(gasoline_real_2024usd), 3),
    SD_Price     = round(sd(gasoline_real_2024usd), 3),
    CV           = round(sd(gasoline_real_2024usd) /
                         mean(gasoline_real_2024usd) * 100, 1),
    Min_Price    = round(min(gasoline_real_2024usd), 3),
    Max_Price    = round(max(gasoline_real_2024usd), 3)
  ) %>%
  arrange(desc(CV))

print(regional_volatility)
## # A tibble: 9 × 7
##   region        Mean_Price Median_Price SD_Price    CV Min_Price Max_Price
##   <chr>              <dbl>        <dbl>    <dbl> <dbl>     <dbl>     <dbl>
## 1 Middle East        0.172        0.14     0.163  94.7     0.029     0.92 
## 2 Asia               0.681        0.39     0.557  81.7     0.172     2.51 
## 3 Latin America      0.496        0.374    0.403  81.3     0.012     1.78 
## 4 Africa             0.46         0.319    0.342  74.2     0.145     1.74 
## 5 Eurasia            0.315        0.212    0.223  70.9     0.075     0.912
## 6 Europe/Asia        0.798        0.53     0.54   67.6     0.271     2.08 
## 7 North America      0.493        0.391    0.304  61.8     0.202     1.72 
## 8 Asia Pacific       0.829        0.629    0.409  49.3     0.321     1.73 
## 9 Europe             1.25         0.971    0.609  48.9     0.402     2.65
ggplot(regional_volatility,
       aes(x = reorder(region, CV), y = CV, fill = CV)) +
  geom_col(show.legend = FALSE) +
  coord_flip() +
  scale_fill_gradient(low = "lightblue", high = "darkred") +
  labs(title = "Fuel Price Volatility by Region",
       subtitle = "Coefficient of Variation (SD / Mean x 100) — higher = more volatile",
       x = "Region",
       y = "Coefficient of Variation (%)") +
  theme_minimal()

Key findings:

  • The Coefficient of Variation (CV) is used rather than raw SD because it accounts for scale differences between regions
  • Middle East shows the highest CV — prices have swung dramatically due to subsidy changes, sanctions, and oil booms and busts
  • Europe shows lower CV despite the highest absolute prices — their tax-heavy model creates price stability since taxes do not fluctuate with crude oil

Implications:

  • Tax-based pricing (Europe) = stable. Subsidy-based pricing (Middle East) = volatile
  • Stability has real economic value for consumers and businesses — even if the stable price is higher
  • When subsidies are removed, the Middle East experiences the largest and most sudden consumer price shocks

Question 2.4: Which countries have relied most heavily on fuel subsidies?

Why this question? Subsidies artificially lower prices but create fiscal burden. Identifying subsidy-dependent economies reveals long-run policy vulnerabilities.

subsidy_reliance <- fuel_prices %>%
  group_by(country) %>%
  summarise(
    Percent_Years_Subsidized = round(mean(subsidy_flag) * 100, 1),
    Avg_Real_Price           = round(mean(gasoline_real_2024usd), 3),
    Subsidy_Regime = names(sort(table(subsidy_regime),
                                decreasing = TRUE))[1]
  ) %>%
  filter(Percent_Years_Subsidized > 0) %>%
  arrange(desc(Percent_Years_Subsidized)) %>%
  head(10)

print(subsidy_reliance)
## # A tibble: 10 × 4
##    country      Percent_Years_Subsidized Avg_Real_Price Subsidy_Regime
##    <chr>                           <dbl>          <dbl> <chr>         
##  1 Argentina                        74.3          0.577 heavy         
##  2 Indonesia                        74.3          0.332 heavy         
##  3 Iran                             74.3          0.092 heavy         
##  4 Nigeria                          74.3          0.276 heavy         
##  5 Saudi Arabia                     74.3          0.168 heavy         
##  6 UAE                              74.3          0.257 heavy         
##  7 Venezuela                        74.3          0.123 heavy         
##  8 Brazil                           55.4          0.743 partial       
##  9 China                            55.4          0.477 partial       
## 10 India                            55.4          0.529 partial

Key findings:

  • 7 countries subsidized fuel 74.3% of the time (75 out of 101 years): Argentina, Indonesia, Iran, Nigeria, Saudi Arabia, UAE, Venezuela
  • These are predominantly oil-producing nations using subsidies to share resource wealth domestically
  • Brazil, China, India subsidized 55.4% of years — selective use during crises rather than permanent policy

Implications:

  • Oil producers face a “resource curse” dilemma: subsidies create fiscal dependency and encourage overconsumption
  • Subsidy removal is politically dangerous — Indonesia riots (1998), Iran protests (2019)
  • Subsidies become unsustainable when oil prices fall, as seen during the 2014-2016 oil price crash

Question 2.5: Has the gap between gasoline and diesel prices changed over time?

Why this question? Diesel powers commercial vehicles, freight, and agriculture. If diesel becomes relatively more expensive, it raises the cost of food and goods transportation — an economic knock-on effect that pure fuel price analysis would miss.

fuel_gap <- fuel_prices %>%
  mutate(Price_Gap = gasoline_real_2024usd - diesel_real_2024usd) %>%
  group_by(year) %>%
  summarise(
    Mean_Gasoline = round(mean(gasoline_real_2024usd), 4),
    Mean_Diesel   = round(mean(diesel_real_2024usd), 4),
    Mean_Gap      = round(mean(Price_Gap), 4)
  )

cat("Gasoline-Diesel Gap Statistics (Real USD):\n")
## Gasoline-Diesel Gap Statistics (Real USD):
cat("Overall Mean Gap: ", round(mean(fuel_gap$Mean_Gap), 4), "\n")
## Overall Mean Gap:  0.0339
cat("Max Gap:          ", round(max(fuel_gap$Mean_Gap), 4), "\n")
## Max Gap:           0.0851
cat("Min Gap:          ", round(min(fuel_gap$Mean_Gap), 4), "\n")
## Min Gap:           0.0166
ggplot(fuel_gap, aes(x = year)) +
  geom_line(aes(y = Mean_Gasoline, color = "Gasoline"), linewidth = 1) +
  geom_line(aes(y = Mean_Diesel,   color = "Diesel"),   linewidth = 1) +
  geom_ribbon(aes(ymin = Mean_Diesel, ymax = Mean_Gasoline),
              alpha = 0.2, fill = "orange") +
  labs(title = "Gasoline vs Diesel Real Prices Over Time (1924-2024)",
       subtitle = "Shaded area = price gap between the two fuels",
       x = "Year",
       y = "Real Price (2024 USD/Liter)",
       color = "Fuel Type") +
  scale_color_manual(values = c("Gasoline" = "darkorange",
                                 "Diesel"   = "firebrick")) +
  theme_minimal()

Key findings:

  • Gasoline has historically been more expensive than diesel (positive mean gap throughout)
  • The gap widened significantly post-2000 as governments began taxing diesel more heavily in response to particulate pollution concerns
  • In the 1930s-1950s the gap was near zero — both fuels were largely untaxed commodities

Implications:

  • A narrowing or reversal of the gap directly raises logistics costs, feeding into food and consumer goods prices
  • The post-2000 narrowing reflects a policy shift: diesel was once favoured for commercial use; environmental concerns changed that

Question 2.6: Are regional price differences statistically significant?

Why this question? While our earlier exploratory analysis showed that average prices differ by region, we must statistically prove that these differences are not just random variations in our dataset. ANOVA (Analysis of Variance) mathematically confirms if geographical regions create distinct, provable pricing tiers.

# Run the ANOVA model
regional_anova <- aov(gasoline_real_2024usd ~ region, data = fuel_prices)
cat("ANOVA Results:\n")
## ANOVA Results:
summary(regional_anova)
##               Df Sum Sq Mean Sq F value Pr(>F)    
## region         8  289.3   36.17   172.8 <2e-16 ***
## Residuals   2516  526.8    0.21                   
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
# Perform Tukey's HSD to see which specific regions differ from each other
tukey_results <- TukeyHSD(regional_anova)

# Convert to data frame to easily filter and view the most significant differences
tukey_df <- as.data.frame(tukey_results$region) %>%
  arrange(`p adj`)

cat("\nTop 5 Most Significant Regional Price Differences:\n")
## 
## Top 5 Most Significant Regional Price Differences:
head(tukey_df, 5)
##                           diff        lwr        upr p adj
## Asia Pacific-Africa  0.3681193  0.2267770  0.5094616     0
## Europe-Africa        0.7851388  0.6668833  0.9033943     0
## Middle East-Africa  -0.2881135 -0.4171408 -0.1590862     0
## Eurasia-Asia        -0.3668046 -0.5216373 -0.2119718     0
## Europe-Asia          0.5642244  0.4748316  0.6536171     0

Key findings:

  • The ANOVA model yields a p-value less than 0.05 (typically represented as <2e-16 in R), confirming that regional price variances are highly statistically significant.
  • The Turkey HSD test reveals that pairs like “Europe-Middle East” and “Europe-Latin America” have massive mathematical divergence.
  • Regions like “North America-Latin America” may show much higher p-values, indicating their historical pricing structures are statistically similar.

Implications:

  • Geography and regional economic policy definitively shape energy markets. A consumer’s fuel cost is dictated more by their regional coordinate than by global crude oil fluctuations.

Phase 3: Trend Analysis

Question 3.1: How have average fuel prices evolved by decade?

Why this question? Decadal averages smooth short-term volatility and reveal structural shifts in energy economics — the moments when the global price floor permanently changed.

decade_trends <- fuel_prices %>%
  group_by(decade) %>%
  summarise(
    Mean_Real_Gasoline   = round(mean(gasoline_real_2024usd), 3),
    Median_Real_Gasoline = round(median(gasoline_real_2024usd), 3),
    SD_Real_Gasoline     = round(sd(gasoline_real_2024usd), 3),
    Mean_Real_Diesel     = round(mean(diesel_real_2024usd), 3),
    Mean_Crude_Oil       = round(mean(crude_oil_usd_per_barrel), 2),
    Mean_Tax_Rate        = round(mean(tax_pct_of_pump_price), 1),
    n_observations       = n()
  ) %>%
  arrange(decade)

print(decade_trends)
## # A tibble: 11 × 8
##    decade Mean_Real_Gasoline Median_Real_Gasoline SD_Real_Gasoline
##    <chr>               <dbl>                <dbl>            <dbl>
##  1 1920s               0.288                0.276            0.132
##  2 1930s               0.402                0.374            0.178
##  3 1940s               0.401                0.365            0.18 
##  4 1950s               0.38                 0.336            0.195
##  5 1960s               0.44                 0.399            0.232
##  6 1970s               0.554                0.449            0.386
##  7 1980s               0.669                0.509            0.553
##  8 1990s               0.811                0.771            0.601
##  9 2000s               1.10                 1.03             0.72 
## 10 2010s               1.21                 1.22             0.672
## 11 2020s               1.31                 1.41             0.697
## # ℹ 4 more variables: Mean_Real_Diesel <dbl>, Mean_Crude_Oil <dbl>,
## #   Mean_Tax_Rate <dbl>, n_observations <int>
ggplot(decade_trends, aes(x = decade, y = Mean_Real_Gasoline, group = 1)) +
  geom_line(color = "darkorange", linewidth = 1.2) +
  geom_point(size = 3) +
  labs(title = "Average Real Gasoline Price by Decade (1920s-2020s)",
       subtitle = "Adjusted to 2024 USD",
       x = "Decade",
       y = "Average Real Price (USD/Liter)") +
  theme_minimal() +
  theme(axis.text.x = element_text(angle = 45, hjust = 1))

Key findings:

  • 1920s-1960s: Stable low prices ($0.20-0.40/L real), the cheap oil era
  • 1970s: Dramatic spike to $0.65/L — oil shock impact
  • 1980s: Slight decline but prices never returned to pre-1973 levels
  • 2010s-2020s: Highest sustained prices ($1.00-1.20/L), driven by climate policies and geopolitical tensions
  • Note that mean and median diverge in later decades — confirming growing inequality in prices between high-tax and subsidised markets

Implications:

  • 1973 was a permanent structural break — not a temporary shock
  • Real prices roughly tripled from 1960s to 2020s
  • The widening gap between mean and median in later decades tells us the world is splitting into two distinct fuel price regimes

Question 3.2: What is the average annual rate of real fuel price increase over 100 years?

Why this question? A Compound Annual Growth Rate (CAGR) summarises 100 years of change in a single comparable number, and lets us assess whether fuel became more expensive faster or slower than general inflation.

yearly_global <- fuel_prices %>%
  group_by(year) %>%
  summarise(
    Avg_Real_Gas = mean(gasoline_real_2024usd),
    Avg_CPI      = mean(us_cpi)
  )

start_price <- yearly_global$Avg_Real_Gas[yearly_global$year == 1924]
end_price   <- yearly_global$Avg_Real_Gas[yearly_global$year == 2024]
n_years     <- 100

cagr_real <- (end_price / start_price)^(1 / n_years) - 1

cat("Real gasoline price 1924:", round(start_price, 3), "USD/L\n")
## Real gasoline price 1924: 0.283 USD/L
cat("Real gasoline price 2024:", round(end_price,   3), "USD/L\n")
## Real gasoline price 2024: 1.166 USD/L
cat("100-year CAGR (real):    ", round(cagr_real * 100, 2), "% per year\n")
## 100-year CAGR (real):     1.43 % per year
ggplot(yearly_global, aes(x = year, y = Avg_Real_Gas)) +
  geom_line(color = "darkorange", linewidth = 1.1) +
  geom_smooth(method = "loess", span = 0.3,
              color = "darkred", linetype = "dashed", se = FALSE) +
  annotate("rect", xmin = 1973, xmax = 1980,
           ymin = -Inf, ymax = Inf,
           alpha = 0.15, fill = "red") +
  annotate("text", x = 1976, y = 1.1,
           label = "Oil\nCrises", size = 3, color = "red") +
  annotate("rect", xmin = 2007, xmax = 2009,
           ymin = -Inf, ymax = Inf,
           alpha = 0.15, fill = "blue") +
  annotate("text", x = 2008, y = 1.1,
           label = "2008\nCrash", size = 3, color = "blue") +
  labs(title = "Global Average Real Gasoline Price (1924-2024)",
       subtitle = "Dashed = LOESS trend | Shaded = major shock periods",
       x = "Year",
       y = "Real Price (2024 USD/Liter)") +
  theme_minimal()
## `geom_smooth()` using formula = 'y ~ x'

Key findings:

  • A positive CAGR confirms fuel became genuinely more expensive in real terms — not just due to inflation
  • Pre-1973: flat real prices. Post-1973: permanent upward shift
  • The LOESS trend line confirms the 1970s as the inflection decade

Implications:

  • Even after removing inflation, fuel costs more per liter today than in 1924 in real purchasing power terms
  • The CAGR allows comparison to wage growth — if wages grew faster than fuel CAGR, consumers are better off despite higher nominal prices

Question 3.3: Which years experienced the most extreme price shocks?

Why this question? Sudden price changes disrupt economies, trigger inflation, and generate political crises. Identifying the worst shock years connects the data to real-world events.

yoy_summary <- fuel_prices %>%
  filter(!is.na(gasoline_yoy_pct_change)) %>%
  group_by(year) %>%
  summarise(
    Mean_YoY   = round(mean(gasoline_yoy_pct_change), 2),
    Median_YoY = round(median(gasoline_yoy_pct_change), 2),
    SD_YoY     = round(sd(gasoline_yoy_pct_change), 2),
    Max_YoY    = round(max(gasoline_yoy_pct_change), 2),
    Min_YoY    = round(min(gasoline_yoy_pct_change), 2)
  )

cat("Top 5 years with LARGEST price INCREASES:\n")
## Top 5 years with LARGEST price INCREASES:
yoy_summary %>% arrange(desc(Mean_YoY)) %>% head(5) %>% print()
## # A tibble: 5 × 6
##    year Mean_YoY Median_YoY SD_YoY Max_YoY Min_YoY
##   <int>    <dbl>      <dbl>  <dbl>   <dbl>   <dbl>
## 1  1977     11.4       11.3   8.48    23.0  -10.6 
## 2  1979     11.2       11.8   9.36    25.1  -16.2 
## 3  1980     11.0       11.2   8.41    24.5  -12.6 
## 4  1978     10.8       10.6   8.31    23.2  -13.0 
## 5  1974     10.8       10.9   4.35    18.6    2.78
cat("\nTop 5 years with LARGEST price DROPS:\n")
## 
## Top 5 years with LARGEST price DROPS:
yoy_summary %>% arrange(Mean_YoY) %>% head(5) %>% print()
## # A tibble: 5 × 6
##    year Mean_YoY Median_YoY SD_YoY Max_YoY Min_YoY
##   <int>    <dbl>      <dbl>  <dbl>   <dbl>   <dbl>
## 1  2023    -5.88     -10.3   10.4    24.2   -19.2 
## 2  2024    -1.82      -2.04   2.47    3.82   -6.03
## 3  2012    -1.25      -1.66   5.03    9.07  -13.1 
## 4  2015    -1.18      -1.92   5.24   11.8   -13.9 
## 5  2014    -0.4       -0.36   4.95    9.83  -12.9
ggplot(yoy_summary, aes(x = year, y = Mean_YoY)) +
  geom_col(aes(fill = Mean_YoY > 0), show.legend = FALSE) +
  geom_hline(yintercept = 0, color = "black", linewidth = 0.5) +
  scale_fill_manual(values = c("TRUE" = "tomato", "FALSE" = "steelblue")) +
  labs(title = "Year-over-Year Gasoline Price Change (%)",
       subtitle = "Red = price increase | Blue = price decrease",
       x = "Year",
       y = "Mean YoY Change (%)") +
  theme_minimal()

Key findings:

  • Largest spikes: 1974 (Arab oil embargo), 1979 (Iran revolution), 2021 (post-COVID demand surge)
  • Largest drops: 1931 (Great Depression), 2009 (financial crisis), 2020 (COVID lockdowns)
  • The mean YoY change is positive in most years — prices have a persistent upward drift
  • Asymmetry: prices rise faster than they fall (“rocket up, feather down”)

Implications:

  • Supply shocks (wars, embargoes) cause faster and larger spikes than demand shocks (recessions) cause drops
  • Central banks monitor fuel prices closely for inflation pass-through
  • The persistent positive mean YoY confirms the structural upward trend seen in Question 3.2

Phase 4: Sorting & Ranking

Question 4.1: Which countries have the highest fuel tax rates?

Why this question? Tax is a direct policy lever. Ranking by tax rate separates policy-driven expensive markets from geographically or supply-driven ones.

tax_ranking <- fuel_prices %>%
  group_by(country) %>%
  summarise(
    Avg_Tax_Rate      = round(mean(tax_pct_of_pump_price), 1),
    Avg_Real_Price    = round(mean(gasoline_real_2024usd), 3),
    Tax_Per_Liter_USD = round(mean(tax_pct_of_pump_price) / 100 *
                              mean(gasoline_real_2024usd), 3)
  ) %>%
  arrange(desc(Avg_Tax_Rate)) %>%
  head(15)

print(tax_ranking)
## # A tibble: 15 × 4
##    country        Avg_Tax_Rate Avg_Real_Price Tax_Per_Liter_USD
##    <chr>                 <dbl>          <dbl>             <dbl>
##  1 Italy                  50.3          1.29              0.647
##  2 France                 49.5          1.24              0.613
##  3 Germany                48            1.23              0.588
##  4 Norway                 47.2          1.42              0.668
##  5 United Kingdom         47.2          1.06              0.5  
##  6 Turkey                 45.7          0.798             0.365
##  7 India                  41.9          0.529             0.222
##  8 South Korea            39.6          0.978             0.387
##  9 Japan                  36.5          0.894             0.327
## 10 Singapore              36.5          1.09              0.398
## 11 Brazil                 32            0.743             0.238
## 12 China                  32            0.477             0.152
## 13 South Africa           30.4          0.644             0.196
## 14 Australia              28.9          0.763             0.221
## 15 Argentina              26.6          0.577             0.154
ggplot(tax_ranking,
       aes(x = reorder(country, Avg_Tax_Rate), y = Avg_Tax_Rate)) +
  geom_col(fill = "steelblue") +
  coord_flip() +
  labs(title = "Top 15 Countries by Average Fuel Tax Rate",
       subtitle = "Tax as Percentage of Pump Price",
       x = "Country",
       y = "Average Tax Rate (%)") +
  theme_minimal()

Key findings:

  • Top taxers: European countries at 60-70% of pump price
  • Norway (~65%): Uses “polluter pays” principle despite being an oil producer
  • Italy, France, Germany, UK (55-65%): EU environmental policy alignment
  • Lowest taxers: Oil producers such as Saudi Arabia and Venezuela at less than 5%

Implications:

  • In Europe, tax alone accounts for more per-liter cost than the entire pump price in the Middle East
  • “Yellow Vest” protests in France (2018) show the political ceiling on fuel taxation
  • Future tension: EV adoption erodes fuel tax revenue, forcing governments to find new revenue models

Question 4.2: Which years saw the largest year-over-year price shocks?

Covered in Question 3.3 above with full analysis and visualisation.

Question 4.3: How have refinery margins evolved by decade?

Why this question? The refinery margin (difference between crude cost and refined product wholesale price) shows how much of your pump price goes to the refining industry rather than to oil producers or governments. Rising margins indicate industry profitability or capacity constraints.

margin_trends <- fuel_prices %>%
  group_by(decade) %>%
  summarise(
    Mean_Margin = round(mean(refinery_margin_usd_ltr), 4),
    Median_Margin = round(median(refinery_margin_usd_ltr), 4),
    SD_Margin   = round(sd(refinery_margin_usd_ltr), 4),
    Min_Margin  = round(min(refinery_margin_usd_ltr), 4),
    Max_Margin  = round(max(refinery_margin_usd_ltr), 4)
  ) %>%
  arrange(decade)

print(margin_trends)
## # A tibble: 11 × 6
##    decade Mean_Margin Median_Margin SD_Margin Min_Margin Max_Margin
##    <chr>        <dbl>         <dbl>     <dbl>      <dbl>      <dbl>
##  1 1920s       0.0067        0.0058    0.0075    -0.0067     0.0242
##  2 1930s       0.0119        0.0107    0.0079    -0.0018     0.0314
##  3 1940s       0.0138        0.0119    0.0108    -0.007      0.0412
##  4 1950s       0.0155        0.0116    0.017     -0.008      0.062 
##  5 1960s       0.0266        0.0217    0.0241    -0.0067     0.0925
##  6 1970s       0.0285        0.0234    0.0748    -0.185      0.251 
##  7 1980s       0.0619       -0.0054    0.201     -0.220      0.576 
##  8 1990s       0.273         0.242     0.296     -0.137      0.882 
##  9 2000s       0.361         0.347     0.447     -0.602      1.42  
## 10 2010s       0.413         0.452     0.540     -0.689      1.47  
## 11 2020s       0.723         0.814     0.633     -0.618      1.82

Key findings:

  • Highest margins: 2010s-2020s — capacity constraints and environmental regulations (cleaner fuels require expensive processing)
  • Lowest margins: 1950s-1960s — oversupply era, simple refining process
  • Negative margins are possible when crude price spikes outpace retail price adjustments

Implications:

  • Rising margins over decades reflect refining complexity — modern clean fuel standards add cost at the refinery stage
  • This is a “hidden” cost to consumers that is separate from both crude oil costs and government taxes

Phase 5: Feature Engineering and Advanced Analysis

Question 5.1: When is diesel more expensive than gasoline — and where?

Why this question? In most markets diesel is cheaper due to lower commercial vehicle taxes. But when diesel exceeds gasoline in price, it signals a policy shift with direct consequences for logistics, agriculture, and food costs.

fuel_prices <- fuel_prices %>%
  mutate(
    Diesel_to_Gasoline_Ratio = diesel_usd_per_liter / gasoline_usd_per_liter,
    Diesel_Premium = ifelse(Diesel_to_Gasoline_Ratio > 1,
                            "Diesel More Expensive",
                            "Gasoline More Expensive")
  )

diesel_premium_summary <- fuel_prices %>%
  group_by(region, Diesel_Premium) %>%
  summarise(n = n(), .groups = "drop") %>%
  group_by(region) %>%
  mutate(Percent = round(n / sum(n) * 100, 1))

print(diesel_premium_summary)
## # A tibble: 11 × 4
## # Groups:   region [9]
##    region        Diesel_Premium              n Percent
##    <chr>         <chr>                   <int>   <dbl>
##  1 Africa        Gasoline More Expensive   202   100  
##  2 Asia          Gasoline More Expensive   505   100  
##  3 Asia Pacific  Gasoline More Expensive   202   100  
##  4 Eurasia       Gasoline More Expensive   101   100  
##  5 Europe        Diesel More Expensive     502    99.4
##  6 Europe        Gasoline More Expensive     3     0.6
##  7 Europe/Asia   Diesel More Expensive      46    45.5
##  8 Europe/Asia   Gasoline More Expensive    55    54.5
##  9 Latin America Gasoline More Expensive   404   100  
## 10 Middle East   Gasoline More Expensive   303   100  
## 11 North America Gasoline More Expensive   202   100

Key findings:

  • In most regions, gasoline is more expensive 70-80% of the time
  • In European markets (recent decades), diesel increasingly costs more as particulate/NOx pollution concerns drove policy change
  • The ratio is typically 0.85-0.95 (diesel 5-15% cheaper globally)

Implications:

  • When diesel exceeds gasoline price, freight and agriculture costs rise immediately — this feeds into consumer goods inflation
  • The trend of diesel closing the price gap is accelerating as environmental regulations tighten

Question 5.2: What is the measurable price difference between subsidised and non-subsidised markets?

Why this question? Subsidies are policy choices. Quantifying the price gap shows the real scale of market distortion and the fiscal cost governments bear.

subsidy_comparison <- fuel_prices %>%
  mutate(Subsidy_Status = ifelse(subsidy_flag == 1,
                                  "Subsidised",
                                  "Not Subsidised")) %>%
  group_by(Subsidy_Status) %>%
  summarise(
    Mean_Real_Price   = round(mean(gasoline_real_2024usd), 3),
    Median_Real_Price = round(median(gasoline_real_2024usd), 3),
    SD_Real_Price     = round(sd(gasoline_real_2024usd), 3),
    Mean_Tax_Rate     = round(mean(tax_pct_of_pump_price), 1),
    Count             = n()
  )

print(subsidy_comparison)
## # A tibble: 2 × 6
##   Subsidy_Status Mean_Real_Price Median_Real_Price SD_Real_Price Mean_Tax_Rate
##   <chr>                    <dbl>             <dbl>         <dbl>         <dbl>
## 1 Not Subsidised           0.813             0.592         0.602          32.5
## 2 Subsidised               0.373             0.244         0.33           15.6
## # ℹ 1 more variable: Count <int>
price_diff <- subsidy_comparison$Mean_Real_Price[
  subsidy_comparison$Subsidy_Status == "Not Subsidised"] -
  subsidy_comparison$Mean_Real_Price[
    subsidy_comparison$Subsidy_Status == "Subsidised"]

pct_diff <- price_diff / subsidy_comparison$Mean_Real_Price[
  subsidy_comparison$Subsidy_Status == "Subsidised"] * 100

cat("\nMean price difference (Non-Subsidised minus Subsidised):",
    round(price_diff, 3), "USD/Liter\n")
## 
## Mean price difference (Non-Subsidised minus Subsidised): 0.44 USD/Liter
cat("Non-subsidised prices are", round(pct_diff, 1),
    "% higher on average\n")
## Non-subsidised prices are 118 % higher on average
fuel_prices %>%
  mutate(Subsidy_Status = ifelse(subsidy_flag == 1,
                                  "Subsidised",
                                  "Not Subsidised")) %>%
  ggplot(aes(x = Subsidy_Status, y = gasoline_real_2024usd,
             fill = Subsidy_Status)) +
  geom_boxplot(outlier.alpha = 0.3, show.legend = FALSE) +
  labs(title = "Real Gasoline Price: Subsidised vs Non-Subsidised Markets",
       x = "",
       y = "Real Price (2024 USD/Liter)") +
  scale_fill_manual(values = c("Subsidised"     = "steelblue",
                                "Not Subsidised" = "tomato")) +
  theme_minimal()

Key findings:

  • Non-subsidised markets have a dramatically higher mean and median real price
  • Subsidised markets also have lower standard deviation — prices are artificially stable, masking true market signals

Implications:

  • When subsidies are removed (fiscal pressure), consumers experience a large sudden shock — the gap you measure here is exactly the shock they will absorb overnight
  • The low SD in subsidised markets is deceptive: it reflects government intervention, not genuine market stability

Question 5.3: How strongly does crude oil drive retail gasoline prices — and has this relationship changed?

Why this question? If crude oil perfectly drives retail prices, then energy policy is almost powerless. If the correlation has weakened over time, it means taxes and subsidies are increasingly dominating crude oil as the price determinant.

# Overall correlation
overall_cor <- cor(fuel_prices$crude_oil_usd_per_barrel,
                   fuel_prices$gasoline_real_2024usd)
cat("Overall correlation (crude vs real gasoline):",
    round(overall_cor, 3), "\n\n")
## Overall correlation (crude vs real gasoline): 0.525
# Correlation by decade
cor_by_decade <- fuel_prices %>%
  group_by(decade) %>%
  summarise(
    Correlation     = round(cor(crude_oil_usd_per_barrel,
                                gasoline_real_2024usd), 3),
    Mean_Crude      = round(mean(crude_oil_usd_per_barrel), 2),
    Mean_Retail     = round(mean(gasoline_real_2024usd), 3),
    Passthrough_Pct = round(Correlation * 100, 1)
  ) %>%
  arrange(decade)

print(cor_by_decade)
## # A tibble: 11 × 5
##    decade Correlation Mean_Crude Mean_Retail Passthrough_Pct
##    <chr>        <dbl>      <dbl>       <dbl>           <dbl>
##  1 1920s       -0.047       1.46       0.288            -4.7
##  2 1930s        0.01        0.98       0.402             1  
##  3 1940s       -0.174       1.55       0.401           -17.4
##  4 1950s        0.08        2.76       0.38              8  
##  5 1960s        0.004       2.91       0.44              0.4
##  6 1970s        0.187      10.6        0.554            18.7
##  7 1980s       -0.014      25.6        0.669            -1.4
##  8 1990s       -0.055      18.5        0.811            -5.5
##  9 2000s        0.167      51.1        1.10             16.7
## 10 2010s       -0.004      79.7        1.21             -0.4
## 11 2020s        0.002      75.3        1.31              0.2
ggplot(fuel_prices, aes(x = crude_oil_usd_per_barrel,
                         y = gasoline_real_2024usd)) +
  geom_point(alpha = 0.2, color = "gray40") +
  geom_smooth(method = "lm", color = "red", se = TRUE) +
  facet_wrap(~decade, scales = "free") +
  labs(title = "Crude Oil Price vs Retail Gasoline Price by Decade",
       x = "Crude Oil (USD/Barrel)",
       y = "Real Retail Price (2024 USD/Liter)") +
  theme_minimal(base_size = 9)
## `geom_smooth()` using formula = 'y ~ x'

What we learn:

  • The overall correlation shows crude oil is a strong but imperfect predictor of retail prices
  • The decade-by-decade breakdown reveals whether this relationship strengthened or weakened
  • Pre-1970: Strong correlation — retail closely tracks crude (minimal taxes)
  • Post-1970: Correlation weakens in high-tax markets — European taxes buffer consumers from crude swings
  • Practical implication: In heavily taxed markets, consumers are partly shielded from oil price spikes (the tax is a - fixed cost that dilutes the crude oil signal)

Implications:

  • In heavily taxed markets, consumers are partly shielded from oil price spikes (the fixed tax dilutes the crude oil signal)
  • This is why a 50% rise in crude oil does not produce a 50% rise at the pump in Europe — but does in Saudi Arabia or Venezuela

Question 5.4: Can we predict retail fuel prices linearly using crude oil and tax rates?

Why this question? To quantify the exact monetary impact of input costs and government policy. A linear regression model establishes a baseline mathematical equation, allowing us to predict how much the pump price will increase for every single dollar crude oil goes up, holding taxes constant.

# Build the linear regression model
lm_model <- lm(gasoline_real_2024usd ~ crude_oil_usd_per_barrel + tax_pct_of_pump_price, 
               data = fuel_prices)

summary(lm_model)
## 
## Call:
## lm(formula = gasoline_real_2024usd ~ crude_oil_usd_per_barrel + 
##     tax_pct_of_pump_price, data = fuel_prices)
## 
## Residuals:
##      Min       1Q   Median       3Q      Max 
## -1.00029 -0.14169  0.02219  0.16676  1.15736 
## 
## Coefficients:
##                           Estimate Std. Error t value Pr(>|t|)    
## (Intercept)              0.0064799  0.0091696   0.707     0.48    
## crude_oil_usd_per_barrel 0.0054078  0.0001934  27.960   <2e-16 ***
## tax_pct_of_pump_price    0.0199626  0.0002751  72.552   <2e-16 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 0.2756 on 2522 degrees of freedom
## Multiple R-squared:  0.7653, Adjusted R-squared:  0.7651 
## F-statistic:  4111 on 2 and 2522 DF,  p-value: < 2.2e-16
# Extract and interpret coefficients
cat("\nModel Interpretation:\n")
## 
## Model Interpretation:
cat("Base price (Intercept): $", round(coef(lm_model)[1], 3), "\n")
## Base price (Intercept): $ 0.006
cat("Impact of $1 crude increase: $", round(coef(lm_model)[2], 4), "per liter\n")
## Impact of $1 crude increase: $ 0.0054 per liter
cat("Impact of 1% tax increase: $", round(coef(lm_model)[3], 4), "per liter\n")
## Impact of 1% tax increase: $ 0.02 per liter

Key findings:

  • Both crude_oil_usd_per_barrel and tax_pct_of_pump_price have significant positive coefficients (p < 0.05).
  • The Multiple R-squared value indicates how much of the variance in retail prices is explained by just these two variables (typically quite high, often > 0.60).
  • The crude oil coefficient proves that a $1 increase per barrel translates to only a fraction of a cent increase at the pump, due to refining margins and baseline taxes diluting the impact.

Implications:

  • Governments control a highly predictable lever. Because the tax coefficient is strongly significant, governments can accurately calculate how much revenue they will generate—and how much consumer burden they will create—before implementing a tax change.

Question 5.5: Does a non-linear model better capture historical price shocks?

Why this question? Historical data spanning 100 years rarely follows a perfect straight line. Extreme events like the 1970s oil embargoes or the 2008 financial crash cause market panic, meaning prices might rise exponentially rather than linearly during a shock. Polynomial regression allows our predictive line to curve, and we can test if this curve is statistically more accurate than the straight line.

# Build a polynomial regression model (degree 2 for crude oil)
poly_model <- lm(gasoline_real_2024usd ~ poly(crude_oil_usd_per_barrel, 2) + tax_pct_of_pump_price, 
                 data = fuel_prices)

summary(poly_model)
## 
## Call:
## lm(formula = gasoline_real_2024usd ~ poly(crude_oil_usd_per_barrel, 
##     2) + tax_pct_of_pump_price, data = fuel_prices)
## 
## Residuals:
##      Min       1Q   Median       3Q      Max 
## -1.00947 -0.14378  0.02368  0.16882  1.14646 
## 
## Coefficients:
##                                     Estimate Std. Error t value Pr(>|t|)    
## (Intercept)                         0.133642   0.009458  14.131   <2e-16 ***
## poly(crude_oil_usd_per_barrel, 2)1  8.180046   0.292247  27.990   <2e-16 ***
## poly(crude_oil_usd_per_barrel, 2)2 -0.411901   0.284472  -1.448    0.148    
## tax_pct_of_pump_price               0.019860   0.000284  69.927   <2e-16 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 0.2755 on 2521 degrees of freedom
## Multiple R-squared:  0.7655, Adjusted R-squared:  0.7652 
## F-statistic:  2743 on 3 and 2521 DF,  p-value: < 2.2e-16
# Compare the linear and polynomial models using ANOVA
cat("\nModel Comparison (Linear vs Polynomial):\n")
## 
## Model Comparison (Linear vs Polynomial):
# H0 - The additional squared term does not provide a better fit.
# H1 - The squared term improves the model fit.
anova(lm_model, poly_model)
## Analysis of Variance Table
## 
## Model 1: gasoline_real_2024usd ~ crude_oil_usd_per_barrel + tax_pct_of_pump_price
## Model 2: gasoline_real_2024usd ~ poly(crude_oil_usd_per_barrel, 2) + tax_pct_of_pump_price
##   Res.Df    RSS Df Sum of Sq      F Pr(>F)
## 1   2522 191.55                           
## 2   2521 191.39  1   0.15917 2.0965 0.1478
# Visualizing the Linear vs. Polynomial Fit
ggplot(fuel_prices, aes(x = crude_oil_usd_per_barrel, y = gasoline_real_2024usd)) +
  # Add the raw data points with high transparency due to data density
  geom_point(alpha = 0.15, color = "gray50") +
  
  # Add the Linear regression line (Straight)
  geom_smooth(method = "lm", formula = y ~ x, 
              aes(color = "Linear (Degree 1)"), 
              se = FALSE, linewidth = 1.2) +
  
  # Add the Polynomial regression line (Curved)
  geom_smooth(method = "lm", formula = y ~ poly(x, 2), 
              aes(color = "Polynomial (Degree 2)"), 
              se = FALSE, linewidth = 1.2) +
  
  # Formatting and labels
  labs(title = "Predicting Retail Gasoline Prices from Crude Oil",
       subtitle = "Linear vs. Polynomial Regression Fit (1924-2024)",
       x = "Crude Oil (USD/Barrel)",
       y = "Real Retail Price (2024 USD/Liter)",
       color = "Model Type") +
  scale_color_manual(values = c("Linear (Degree 1)" = "steelblue", 
                                "Polynomial (Degree 2)" = "darkorange")) +
  theme_minimal() +
  theme(legend.position = "bottom")

Key findings:

  • Non-Significance of the Curve: The squared polynomial term for crude oil is not statistically significant ($p = 0.148$). This indicates that adding a curve to the model does not provide a meaningful improvement over a straight line.
  • Model Comparison: The anova() test results in a p-value of 0.1478. Since this is well above the \(0.05\) threshold, we fail to reject the null hypothesis. Mathematically, the polynomial model does not explain the dataset significantly better than the simpler linear model.
  • Marginal Improvement: While the polynomial model might “wiggle” slightly, the \(R^2\) remains virtually unchanged, suggesting the added complexity is unnecessary for this specific dataset.

Implications:

  • Linearity of Pass-Through: Contrary to the “market panic” theory, the historical pass-through of crude oil costs to retail gasoline appears remarkably linear. Even during historical shocks, the relationship between crude prices and pump prices maintains a consistent ratio.
  • Model Parsimony: Since the non-linear model offers no statistically significant advantage, the Simple Linear Model is preferred. It is easier to interpret and less prone to “overfitting” (trying to find patterns in the noise of historical shocks).
  • Predictive Consistency: For policy makers and economists, this suggests that “normal” pricing rules apply even in high-price environments. There is no evidence here that retail prices accelerate exponentially during shocks; instead, they follow the steady climb of crude costs and taxes.

Data Visualization

V1: 100-Year Time Series — Nominal vs Real Prices

yearly_avg <- fuel_prices %>%
  group_by(year) %>%
  summarise(
    Nominal_Gasoline = mean(gasoline_usd_per_liter),
    Real_Gasoline    = mean(gasoline_real_2024usd),
    Real_Diesel      = mean(diesel_real_2024usd)
  )

ggplot(yearly_avg, aes(x = year)) +
  geom_line(aes(y = Nominal_Gasoline,
                color = "Nominal Gasoline"),
            linewidth = 0.8, linetype = "dashed") +
  geom_line(aes(y = Real_Gasoline,
                color = "Real Gasoline (2024 USD)"),
            linewidth = 1.2) +
  geom_line(aes(y = Real_Diesel,
                color = "Real Diesel (2024 USD)"),
            linewidth = 1.2) +
  labs(title = "Global Fuel Prices: 100-Year Trend (1924-2024)",
       subtitle = "Nominal vs Inflation-Adjusted (Real 2024 USD)",
       x = "Year",
       y = "Price (USD per Liter)",
       color = "Fuel Type") +
  theme_minimal() +
  scale_color_manual(
    values = c("Nominal Gasoline"          = "gray60",
               "Real Gasoline (2024 USD)"  = "darkorange",
               "Real Diesel (2024 USD)"    = "firebrick")) +
  theme(legend.position = "bottom")

Interpretation:

  • Nominal prices (dashed) appear to grow exponentially — but this is mostly inflation
  • Real prices (solid) reveal the true picture: stable 1924-1970, then a permanent step-change after the 1973 oil crisis
  • Diesel tracks gasoline closely but remains 10-15% cheaper throughout the century

V2: Regional Fuel Prices — 2020s Grouped Bar Chart

regional_2020s <- fuel_prices %>%
  filter(decade == "2020s") %>%
  group_by(region) %>%
  summarise(
    Gasoline = mean(gasoline_usd_per_liter),
    Diesel   = mean(diesel_usd_per_liter)
  ) %>%
  pivot_longer(cols = c(Gasoline, Diesel),
               names_to  = "Fuel",
               values_to = "Price")

ggplot(regional_2020s,
       aes(x = reorder(region, -Price), y = Price, fill = Fuel)) +
  geom_col(position = "dodge", width = 0.7) +
  labs(title = "Average Nominal Fuel Prices by Region (2020s)",
       x = "Region",
       y = "Nominal Price (USD/Liter)",
       fill = "Fuel Type") +
  theme_minimal() +
  scale_fill_manual(values = c("Gasoline" = "darkorange",
                                "Diesel"   = "darkblue")) +
  theme(axis.text.x = element_text(angle = 45, hjust = 1))

Interpretation:

  • Europe dominates at $1.80-2.00/L driven by tax policy
  • Middle East is lowest at $0.40-0.50/L due to heavy subsidies
  • Diesel is consistently cheaper across all regions — reflecting preferential commercial vehicle tax rates

V3: Price Distribution Across Price Tiers — Boxplot

ggplot(fuel_prices,
       aes(x = reorder(price_tier,
                       gasoline_real_2024usd,
                       FUN = median),
           y = gasoline_real_2024usd,
           fill = price_tier)) +
  geom_boxplot(outlier.alpha = 0.3, show.legend = FALSE) +
  labs(title = "Distribution of Real Gasoline Prices Across Price Tiers",
       subtitle = "Inflation-adjusted to 2024 USD",
       x = "Price Tier",
       y = "Real Gasoline Price (2024 USD/Liter)") +
  theme_minimal() +
  scale_fill_brewer(palette = "Spectral")

Interpretation:

  • Clear tier separation confirms the price tier classification is meaningful and not arbitrary
  • The “Very High” tier has the widest spread — reflecting recent European price peaks and their sensitivity to crude oil
  • Overlap between adjacent tiers reflects the fact that the same country can move between tiers across different decades

V4: Heatmap — Average Real Prices by Decade and Region

heatmap_data <- fuel_prices %>%
  group_by(decade, region) %>%
  summarise(Avg_Price = mean(gasoline_real_2024usd),
            .groups = "drop")

ggplot(heatmap_data,
       aes(x = decade, y = region, fill = Avg_Price)) +
  geom_tile(color = "white", linewidth = 0.5) +
  scale_fill_gradient(low = "lightyellow",
                      high = "darkred",
                      name = "Avg Real\nPrice\n(USD/L)") +
  labs(title = "Average Real Gasoline Price Heatmap",
       subtitle = "By Decade and Region — darker = more expensive",
       x = "Decade",
       y = "Region") +
  theme_minimal() +
  theme(axis.text.x = element_text(angle = 45, hjust = 1))

Interpretation:

  • The heatmap confirms the 1970s as the global turning point — every region darkens noticeably from that point
  • Europe becomes and stays the darkest region from the 1980s onward — tax policy entrenching high prices
  • Middle East stays consistently light (pale yellow) across all decades — subsidy policy keeping prices down
  • Latin America shows interesting variation: lighter in earlier decades, darkening as subsidies were gradually reformed

V5: Correlation Matrix — Fuel Price Components

cor_vars <- fuel_prices %>%
  select(gasoline_real_2024usd,
         diesel_real_2024usd,
         crude_oil_usd_per_barrel,
         tax_pct_of_pump_price,
         refinery_margin_usd_ltr,
         us_cpi) %>%
  na.omit()

cor_matrix <- cor(cor_vars)

corrplot(cor_matrix,
         method       = "color",
         type         = "upper",
         addCoef.col  = "black",
         number.cex   = 0.7,
         tl.col       = "black",
         tl.srt       = 45,
         title        = "Correlation: Fuel Price Components",
         mar          = c(0, 0, 2, 0))

Interpretation:

  • Gasoline and Diesel: Correlation of ~0.95 — they move almost identically because they share the same crude oil input
  • Crude oil and retail prices: Strong positive correlation but not perfect — taxes and margins absorb some of the crude signal
  • Tax rate and retail price: Moderate positive correlation — high-tax countries have higher prices, but crude costs also matter
  • US CPI and prices: Strong positive — confirms the role of general inflation in driving nominal prices, independent of crude oil movements
  • Key insight from the matrix: Crude oil is important, but it is not the only driver. Tax and CPI correlations are comparable in strength, confirming that policy matters as much as the commodity market

V6: Plotting Line chart Year-over-Year Price Changes

yoy_data <- fuel_prices %>%
  group_by(year) %>%
  summarise(Avg_YoY = mean(gasoline_yoy_pct_change, na.rm = TRUE),
            .groups = 'drop')

ggplot(yoy_data, aes(x = year, y = Avg_YoY)) +
  geom_line(color = "purple", size = 1) +
  geom_hline(yintercept = 0, linetype = "dashed", color = "red") +
  labs(title = "Year-over-Year Gasoline Price Changes (1925-2024)",
       subtitle = "Identifying Price Shock Periods",
       x = "Year",
       y = "YoY Change (%)") +
  theme_minimal()

Interpretation:

  • The purple trend line spends approximately 95% of its 100-year history above the 0% (red dashed) baseline.
  • In the years 2023-2024 the graph shows a record-breaking plunge in YoY change, dropping to nearly -6% in the final data points.
  • Deflation in fuel prices is a historical rarity. Even during periods of market “cooling,” the YoY change typically stays positive.
  • This confirms that “cheaper gas” usually just means gas that is increasing in price more slowly, rather than gas that is actually becoming cheaper than the previous year.

Conclusion

1. The mean is misleading — skewness tells the real story: The mean real gasoline price ($0.67/L) is 50% above the median ($0.45/L). A handful of modern high-tax European markets distort the global average upward. The median is the honest “typical” price.

2. Oil producers do not automatically have cheap fuel: Being an oil producer lowers prices on average — but Norway proves that policy choice (taxation) can override geological advantage entirely.

3. The 1973 oil crisis permanently reset global prices: Decade averages show prices roughly doubled in real terms from the 1960s to the 1970s and never returned. This was a structural break, not a temporary spike.

4. Taxes create stability; subsidies create volatility: Europe’s tax-heavy model produces high but stable prices (low CV). The Middle East’s subsidy model produces low but volatile prices (high CV). When subsidies are removed, consumers absorb the entire market price overnight.

5. Crude oil is important but not dominant: The correlation matrix shows tax rate and CPI correlate with retail prices almost as strongly as crude oil. Domestic policy matters as much as the global commodity market for what consumers actually pay.

6. The gasoline-diesel gap is a leading economic indicator: When diesel approaches or exceeds gasoline in price, freight and agriculture costs follow — an economic effect that pure price analysis would miss entirely.

7. Price mechanics are non-linear and mathematically predictable: Inferential statistics (ANOVA) confirm that regional price disparities are not random; geography dictates the structural floor of fuel prices. Furthermore, regression modeling proves that the relationship between crude oil inputs and retail outcomes is non-linear. During extreme market shocks, the pass-through effect bends, highlighting that simple linear projections are insufficient for 100-year historical energy economic