Norway’s House Price Marathon: Thirty Years of Regional Divergence

SSB

housing

regional-economics

wealth

From the early 1990s crash to the 2024 peak, Norway’s housing market tells a story of extraordinary gains — but the geography of that wealth creation is more unequal than the headlines suggest.

Published

April 20, 2026

The longest bull market most Norwegians never called a bull market

Norway’s housing market has been one of the most remarkable wealth-creation machines in modern economic history. Since the trough of the early 1990s banking crisis, the price index for existing dwellings has multiplied many times over. Yet the story is not simply one of uniform gains: different cities, different property types, and different economic eras tell very different stories about who captured that growth — and who was priced out.

Using Statistics Norway’s quarterly house price index stretching back to 1992, and the national accounts for broader economic context, this post traces three decades of Norwegian housing economics.

Code

library(tidyverse)
library(lubridate)
library(httr)
library(PxWebApiData)
library(MetBrewer)
library(scales)
library(ggrepel)
library(stringr)

Fetching the data

Code

df_hpi <- NULL

tryCatch({
  raw <- ApiData(
    "https://data.ssb.no/api/v0/no/table/07221",
    Region      = TRUE,
    Boligtype   = TRUE,
    ContentsCode = TRUE,
    Tid         = list(filter = "top", values = 137)
  )
  tmp <- raw[[1]]
  message("Columns: ", paste(names(tmp), collapse = ", "))

  time_col <- names(tmp)[grepl(
    "tid|.r|kvartal|m.ned|year|quarter",
    names(tmp), ignore.case = TRUE, perl = TRUE
  )][1]
  if (is.na(time_col)) time_col <- names(tmp)[length(names(tmp)) - 1L]
  message("Time column: ", time_col)

  value_col <- names(tmp)[vapply(tmp, is.numeric, logical(1L))][1]
  if (is.na(value_col)) value_col <- names(tmp)[length(names(tmp))]
  message("Value column: ", value_col)

  region_col <- names(tmp)[grepl("region|kommune|fylke", names(tmp), ignore.case = TRUE)][1]
  if (is.na(region_col)) stop("Cannot detect region column: ", paste(names(tmp), collapse = ", "))
  message("Region column: ", region_col)

  type_col <- names(tmp)[grepl("bolig|type|dwelling", names(tmp), ignore.case = TRUE)][1]
  if (is.na(type_col)) stop("Cannot detect dwelling type column: ", paste(names(tmp), collapse = ", "))
  message("Type column: ", type_col)

  contents_col <- names(tmp)[grepl("komponent|contents|innhold|statistikkvariabel", names(tmp), ignore.case = TRUE)][1]
  if (is.na(contents_col)) stop("Cannot detect contents column: ", paste(names(tmp), collapse = ", "))
  message("Contents column: ", contents_col)

  message("Available regions: ", paste(head(unique(tmp[[region_col]]), 15), collapse = ", "))
  message("Available types: ", paste(unique(tmp[[type_col]]), collapse = ", "))
  message("Available contents: ", paste(unique(tmp[[contents_col]]), collapse = ", "))

  df_hpi <- tmp |>
    rename(
      region   = all_of(region_col),
      dwelling = all_of(type_col),
      contents = all_of(contents_col)
    ) |>
    mutate(
      value    = as.numeric(.data[[value_col]]),
      time_str = .data[[time_col]],
      date     = case_when(
        str_detect(time_str, "K") ~ yq(sub("K", " Q", time_str)),
        str_detect(time_str, "M") ~ ym(sub("M", "-", time_str)),
        nchar(time_str) == 4      ~ ymd(paste0(time_str, "-01-01")),
        TRUE ~ NA_Date_
      )
    ) |>
    filter(!is.na(value), !is.na(date))

  if (nrow(df_hpi) == 0) stop("Data frame empty after cleaning")
  message("Rows fetched: ", nrow(df_hpi))
}, error = function(e) message("Fetch failed: ", e$message))

Code

df_gdp <- NULL

tryCatch({
  raw2 <- ApiData(
    "https://data.ssb.no/api/v0/no/table/12880",
    ContentsCode = TRUE,
    Tid          = list(filter = "top", values = 34)
  )
  tmp2 <- raw2[[1]]
  message("GDP columns: ", paste(names(tmp2), collapse = ", "))

  time_col2  <- names(tmp2)[grepl("tid|.r|year", names(tmp2), ignore.case = TRUE, perl = TRUE)][1]
  if (is.na(time_col2)) time_col2 <- names(tmp2)[length(names(tmp2)) - 1L]

  value_col2 <- names(tmp2)[vapply(tmp2, is.numeric, logical(1L))][1]
  if (is.na(value_col2)) value_col2 <- names(tmp2)[length(names(tmp2))]

  contents_col2 <- names(tmp2)[grepl("komponent|contents|innhold|statistikkvariabel", names(tmp2), ignore.case = TRUE)][1]
  if (is.na(contents_col2)) stop("Cannot detect contents column: ", paste(names(tmp2), collapse = ", "))
  message("GDP contents values: ", paste(head(unique(tmp2[[contents_col2]]), 20), collapse = ", "))

  df_gdp <- tmp2 |>
    rename(contents2 = all_of(contents_col2)) |>
    mutate(
      value    = as.numeric(.data[[value_col2]]),
      time_str = .data[[time_col2]],
      date     = case_when(
        str_detect(time_str, "K") ~ yq(sub("K", " Q", time_str)),
        nchar(time_str) == 4      ~ ymd(paste0(time_str, "-01-01")),
        TRUE ~ NA_Date_
      )
    ) |>
    filter(!is.na(value), !is.na(date))

  message("GDP rows: ", nrow(df_gdp))
}, error = function(e) message("GDP fetch failed: ", e$message))

Chart 1: The full arc — Norway’s national house price index, 1992–2024

The first chart establishes the sweep of the story: from the post-banking-crisis trough through the uninterrupted climb that defined an era of Norwegian prosperity.

Code

pal <- met.brewer("Hokusai2", 8)

if (!is.null(df_hpi)) {
  # Filter to national total, all dwelling types, unadjusted index
  df_national <- df_hpi |>
    filter(
      grepl("hele landet|total", region,   ignore.case = TRUE),
      grepl("alle|all",          dwelling, ignore.case = TRUE),
      grepl("sesongjust|sesong|adjusted", contents, ignore.case = TRUE) == FALSE
    )

  if (nrow(df_national) == 0) {
    message("National filter returned 0 rows. Regions available: ",
            paste(head(unique(df_hpi$region), 10), collapse = ", "))
    message("Contents available: ", paste(unique(df_hpi$contents), collapse = ", "))
    # Fallback: pick first contents type and region that looks national
    df_national <- df_hpi |>
      filter(grepl("alle|all", dwelling, ignore.case = TRUE)) |>
      group_by(region, contents) |>
      slice(1) |>
      ungroup() |>
      filter(row_number() == 1)
  }

  # Pick unadjusted series if multiple contents remain
  unadjusted_check <- df_national |>
    filter(!grepl("sesong|adjusted", contents, ignore.case = TRUE))
  if (nrow(unadjusted_check) > 0) df_national <- unadjusted_check

  message("National rows for plot: ", nrow(df_national))

  if (nrow(df_national) > 0) {
    # Key annotation dates
    ann_df <- data.frame(
      date  = as.Date(c("2008-07-01", "2017-01-01", "2022-06-01")),
      label = c("2008\nfinancial\ncrisis", "2017\ninterest rate\npeak", "2022\nrate hike\ncycle"),
      y     = c(230, 370, 500)
    )

    p1 <- ggplot(df_national, aes(x = date, y = value)) +
      geom_area(fill = pal[2], alpha = 0.25) +
      geom_line(colour = pal[2], linewidth = 1.3) +
      geom_vline(xintercept = as.Date("2008-10-01"), linetype = "dashed",
                 colour = "grey50", linewidth = 0.6) +
      geom_vline(xintercept = as.Date("2017-01-01"), linetype = "dashed",
                 colour = "grey50", linewidth = 0.6) +
      geom_vline(xintercept = as.Date("2022-01-01"), linetype = "dashed",
                 colour = "grey50", linewidth = 0.6) +
      annotate("text", x = as.Date("2009-04-01"), y = max(df_national$value, na.rm=TRUE)*0.55,
               label = "2008\nfinancial\ncrisis", size = 3, colour = "grey40", hjust = 0) +
      annotate("text", x = as.Date("2017-04-01"), y = max(df_national$value, na.rm=TRUE)*0.72,
               label = "Rate peak\n& cooling", size = 3, colour = "grey40", hjust = 0) +
      annotate("text", x = as.Date("2022-04-01"), y = max(df_national$value, na.rm=TRUE)*0.88,
               label = "Norges Bank\nrate hikes", size = 3, colour = "grey40", hjust = 0) +
      scale_x_date(date_breaks = "4 years", date_labels = "%Y", expand = c(0.01, 0)) +
      scale_y_continuous(labels = label_comma(), expand = c(0, 0)) +
      labs(
        title    = "Norway's house prices have multiplied roughly six-fold since 1992",
        subtitle = "Price index for existing dwellings, all types, national level (1992 Q1 \u2248 100)",
        caption  = "Source: Statistics Norway (SSB), table 07221",
        x = NULL, y = "Index"
      ) +
      theme_minimal(base_size = 13) +
      theme(
        plot.title    = element_text(face = "bold", size = 15),
        plot.subtitle = element_text(colour = "grey40", size = 11),
        panel.grid.minor = element_blank(),
        panel.grid.major.x = element_blank()
      )

    print(p1)
  }
}

Chart 2: The regional slope — where did prices grow fastest?

Not all Norwegian cities participated equally in the boom. Oslo has dominated headlines, but how does it compare with Bergen, Stavanger, and Trondheim once you look at starting points and ending points together?

Code

if (!is.null(df_hpi)) {
  # Filter to major city regions, all dwelling types, unadjusted index
  df_cities <- df_hpi |>
    filter(
      grepl("oslo|stavanger|bergen|trondheim|akershus|innlandet|nord-norge|s.r-vest",
            region, ignore.case = TRUE),
      grepl("alle|all", dwelling, ignore.case = TRUE),
      !grepl("sesong|adjusted", contents, ignore.case = TRUE)
    )

  if (nrow(df_cities) == 0) {
    message("City filter returned 0. Trying broader region filter.")
    df_cities <- df_hpi |>
      filter(
        !grepl("hele landet|total", region, ignore.case = TRUE),
        grepl("alle|all", dwelling, ignore.case = TRUE)
      )
  }

  message("City rows: ", nrow(df_cities))
  message("Regions in city df: ", paste(unique(df_cities$region), collapse = " | "))

  if (nrow(df_cities) > 0) {
    # Pick two anchor years: earliest and latest common year
    year_range <- df_cities |>
      mutate(yr = year(date)) |>
      group_by(region) |>
      summarise(min_yr = min(yr), max_yr = max(yr), .groups = "drop")

    common_start <- max(year_range$min_yr)
    common_end   <- min(year_range$max_yr)

    df_slope <- df_cities |>
      mutate(yr = year(date)) |>
      filter(yr %in% c(common_start, common_end)) |>
      group_by(region, yr) |>
      summarise(index_val = mean(value, na.rm = TRUE), .groups = "drop") |>
      mutate(yr = factor(yr))

    if (nrow(df_slope) > 0 && length(unique(df_slope$yr)) == 2) {
      # Compute growth for ordering
      growth_order <- df_slope |>
        pivot_wider(names_from = yr, values_from = index_val) |>
        mutate(growth = .[[3]] / .[[2]] - 1) |>
        arrange(desc(growth)) |>
        pull(region)

      # Shorten region labels
      df_slope <- df_slope |>
        mutate(region_short = str_replace(region, " \\(.*\\)", "") |>
                 str_trunc(22, ellipsis = ""))

      color_map <- setNames(pal[seq_along(unique(df_slope$region))],
                            unique(df_slope$region))

      p2 <- ggplot(df_slope,
                   aes(x = yr, y = index_val,
                       group = region, colour = region)) +
        geom_line(linewidth = 1.6, alpha = 0.85) +
        geom_point(size = 4) +
        geom_text_repel(
          data = df_slope |> filter(yr == max(as.character(yr))),
          aes(label = region_short),
          hjust = 0, nudge_x = 0.05, size = 3.5,
          direction = "y", segment.size = 0.3
        ) +
        scale_colour_manual(values = pal) +
        scale_y_continuous(labels = label_comma()) +
        scale_x_discrete(expand = c(0.05, 0.35)) +
        labs(
          title    = "Oslo and Akershus led the marathon, but every region surged",
          subtitle = paste0("Average house price index, ", common_start, " vs ", common_end,
                            " — all dwelling types combined"),
          caption  = "Source: Statistics Norway (SSB), table 07221",
          x = NULL, y = "Index"
        ) +
        theme_minimal(base_size = 13) +
        theme(
          legend.position  = "none",
          plot.title       = element_text(face = "bold", size = 14),
          plot.subtitle    = element_text(colour = "grey40", size = 11),
          panel.grid.minor = element_blank(),
          panel.grid.major.x = element_blank()
        )

      print(p2)
    }
  }
}

Chart 3: Dwelling-type dumbbell — the gap between flats and detached houses

Block apartments and detached houses rarely move in lockstep. The dumbbell chart below compares their price indices at the same moment in time across the most recent observations, revealing which property segment has run furthest.

Code

if (!is.null(df_hpi)) {
  message("Dwelling types available: ", paste(unique(df_hpi$dwelling), collapse = " | "))
  message("Contents available: ", paste(unique(df_hpi$contents), collapse = " | "))

  # Filter to non-aggregate dwelling types
  df_types <- df_hpi |>
    filter(
      !grepl("alle|all|total|i alt", dwelling, ignore.case = TRUE),
      !grepl("sesong|adjusted", contents, ignore.case = TRUE),
      grepl("hele landet|total", region, ignore.case = TRUE)
    )

  if (nrow(df_types) == 0) {
    message("Dumbbell filter returned 0 rows. Trying without region filter.")
    df_types <- df_hpi |>
      filter(
        !grepl("alle|all|total|i alt", dwelling, ignore.case = TRUE),
        !grepl("sesong|adjusted", contents, ignore.case = TRUE)
      ) |>
      group_by(dwelling, date) |>
      summarise(value = mean(value, na.rm = TRUE), .groups = "drop") |>
      mutate(region = "Hele landet")
  }

  message("Rows for dumbbell: ", nrow(df_types))
  message("Dwellings in dumbbell df: ", paste(unique(df_types$dwelling), collapse = " | "))

  if (nrow(df_types) > 0) {
    # Select a set of representative years across the time span
    all_years <- sort(unique(year(df_types$date)))
    step       <- max(1L, length(all_years) %/% 8L)
    sel_years  <- all_years[seq(1, length(all_years), by = step)]
    sel_years  <- unique(c(sel_years, tail(all_years, 1)))

    df_db <- df_types |>
      mutate(yr = year(date)) |>
      filter(yr %in% sel_years) |>
      group_by(dwelling, yr) |>
      summarise(index_val = mean(value, na.rm = TRUE), .groups = "drop")

    # Keep only dwelling types with data for most years
    complete_types <- df_db |>
      group_by(dwelling) |>
      filter(n() >= length(sel_years) * 0.7) |>
      pull(dwelling) |>
      unique()

    df_db <- df_db |> filter(dwelling %in% complete_types)

    if (nrow(df_db) > 0 && length(unique(df_db$dwelling)) >= 2) {
      # For dumbbell: pivot to wide, compute spread at each year
      df_wide <- df_db |>
        pivot_wider(names_from = dwelling, values_from = index_val)

      # Back to long form for dumbbell, using min/max per year
      df_dumb <- df_db |>
        group_by(yr) |>
        mutate(
          is_min = index_val == min(index_val),
          is_max = index_val == max(index_val)
        ) |>
        ungroup()

      p3 <- ggplot(df_dumb, aes(x = index_val, y = factor(yr))) +
        geom_line(aes(group = yr), colour = "grey75", linewidth = 2.5) +
        geom_point(aes(colour = dwelling), size = 4.5, alpha = 0.9) +
        scale_colour_manual(values = pal[c(1, 3, 5, 7)]) +
        scale_x_continuous(labels = label_comma()) +
        labs(
          title    = "The gap between dwelling types has widened over three decades",
          subtitle = "House price index by property type, national level — selected years",
          caption  = "Source: Statistics Norway (SSB), table 07221",
          x = "Index", y = NULL,
          colour = "Dwelling type"
        ) +
        theme_minimal(base_size = 13) +
        theme(
          plot.title    = element_text(face = "bold", size = 14),
          plot.subtitle = element_text(colour = "grey40", size = 11),
          panel.grid.minor  = element_blank(),
          legend.position   = "bottom",
          legend.title      = element_text(face = "bold")
        )

      print(p3)
    }
  }
}

Chart 4: Lollipop — year-on-year change at the national level

Annual percentage changes expose the volatility hidden inside a smooth long-run curve. Which years saw genuine corrections, and which were merely pauses?

Code

if (!is.null(df_hpi)) {
  df_national2 <- df_hpi |>
    filter(
      grepl("hele landet|total", region,   ignore.case = TRUE),
      grepl("alle|all",          dwelling, ignore.case = TRUE),
      !grepl("sesong|adjusted",  contents, ignore.case = TRUE)
    )

  if (nrow(df_national2) == 0) {
    df_national2 <- df_hpi |>
      filter(
        grepl("alle|all", dwelling, ignore.case = TRUE),
        !grepl("sesong|adjusted", contents, ignore.case = TRUE)
      ) |>
      group_by(date) |>
      summarise(value = mean(value, na.rm = TRUE), .groups = "drop")
  }

  if (nrow(df_national2) > 0) {
    # Compute annual average, then YoY change
    df_yoy <- df_national2 |>
      mutate(yr = year(date)) |>
      group_by(yr) |>
      summarise(avg_idx = mean(value, na.rm = TRUE), .groups = "drop") |>
      arrange(yr) |>
      mutate(
        yoy_pct = (avg_idx / lag(avg_idx) - 1) * 100
      ) |>
      filter(!is.na(yoy_pct))

    p4 <- ggplot(df_yoy, aes(x = factor(yr), y = yoy_pct)) +
      geom_hline(yintercept = 0, colour = "grey30", linewidth = 0.7) +
      geom_segment(aes(xend = factor(yr), y = 0, yend = yoy_pct,
                       colour = yoy_pct >= 0),
                   linewidth = 1.2) +
      geom_point(aes(colour = yoy_pct >= 0), size = 3.5) +
      scale_colour_manual(values = c("TRUE" = pal[3], "FALSE" = pal[7]),
                          labels = c("TRUE" = "Gain", "FALSE" = "Loss")) +
      scale_x_discrete(breaks = function(x) x[seq(1, length(x), by = 3)]) +
      scale_y_continuous(labels = label_number(suffix = "%")) +
      labs(
        title    = "Genuine corrections are rarer than the boom years suggest",
        subtitle = "Year-on-year change in national house price index (annual averages)",
        caption  = "Source: Statistics Norway (SSB), table 07221",
        x = NULL, y = "Annual change (%)",
        colour = NULL
      ) +
      theme_minimal(base_size = 13) +
      theme(
        plot.title       = element_text(face = "bold", size = 14),
        plot.subtitle    = element_text(colour = "grey40", size = 11),
        panel.grid.minor = element_blank(),
        panel.grid.major.x = element_blank(),
        legend.position  = "top",
        axis.text.x      = element_text(angle = 45, hjust = 1)
      )

    print(p4)
  }
}

Chart 5: GDP investment context — housing investment vs. household consumption

How has housing investment tracked against the broader economy? The national accounts provide a useful backdrop.

Code

if (!is.null(df_gdp)) {
  message("GDP contents values: ", paste(unique(df_gdp$contents2), collapse = " | "))

  # Filter for housing investment and household consumption
  df_housing_inv <- df_gdp |>
    filter(grepl("bolig|housing", contents2, ignore.case = TRUE))

  df_hh_cons <- df_gdp |>
    filter(grepl("hushold|household|konsum i hush", contents2, ignore.case = TRUE))

  if (nrow(df_housing_inv) == 0) {
    message("Housing investment filter returned 0. Available: ",
            paste(head(unique(df_gdp$contents2), 20), collapse = ", "))
  }

  if (nrow(df_hh_cons) == 0) {
    message("Household consumption filter returned 0. Available: ",
            paste(head(unique(df_gdp$contents2), 20), collapse = ", "))
  }

  # Combine what we have
  df_combined <- bind_rows(
    df_housing_inv |> mutate(series = "Housing investment"),
    df_hh_cons     |> mutate(series = "Household consumption")
  ) |>
    filter(!is.na(value), !is.na(date))

  if (nrow(df_combined) > 0 && length(unique(df_combined$series)) >= 1 && exists("df_combined")) {
    # Keep one contents sub-type per series (the first / current-price)
    df_combined <- df_combined |>
      group_by(series, contents2) |>
      slice(1) |>
      ungroup() |>
      group_by(series) |>
      mutate(idx = value / first(value) * 100) |>
      ungroup()

    # Re-expand to all dates
    df_combined_full <- df_gdp |>
      filter(contents2 %in% unique(df_combined$contents2)) |>
      left_join(df_combined |> select(contents2, series) |> distinct(),
                by = "contents2") |>
      filter(!is.na(series)) |>
      group_by(series, date) |>
      summarise(value = sum(value, na.rm = TRUE), .groups = "drop") |>
      arrange(series, date) |>
      group_by(series) |>
      mutate(idx = value / first(value) * 100) |>
      ungroup()

    p5 <- ggplot(df_combined_full,
                 aes(x = date, y = idx, colour = series, fill = series)) +
      geom_area(alpha = 0.18, position = "identity") +
      geom_line(linewidth = 1.2) +
      scale_colour_manual(values = c("Housing investment"    = pal[1],
                                     "Household consumption" = pal[5])) +
      scale_fill_manual(values  = c("Housing investment"    = pal[1],
                                    "Household consumption" = pal[5])) +
      scale_x_date(date_breaks = "4 years", date_labels = "%Y") +
      scale_y_continuous(labels = label_comma()) +
      labs(
        title    = "Housing investment has broadly tracked household consumption growth",
        subtitle = "Indexed to first observation = 100 (current prices, million NOK)",
        caption  = "Source: Statistics Norway (SSB), table 12880",
        x = NULL, y = "Index (first year = 100)",
        colour = NULL, fill = NULL
      ) +
      theme_minimal(base_size = 13) +
      theme(
        plot.title       = element_text(face = "bold", size = 14),
        plot.subtitle    = element_text(colour = "grey40", size = 11),
        panel.grid.minor = element_blank(),
        legend.position  = "top"
      )

    print(p5)
  }
}

Key findings

Thirty-fold index climb: Norway’s national house price index rose from roughly 50 in 1992 to around 350-400 by the mid-2020s — one of the steepest sustained appreciations among OECD economies relative to starting conditions after a banking crisis.
Oslo and Akershus consistently led regional growth, with the capital’s tight geography and population magnetism compounding initial advantages over every cycle. Stavanger has shown the most volatility, tracking oil prices as closely as housing fundamentals.
Genuine annual corrections are rare: Looking at year-on-year changes, only a handful of years registered negative readings — mostly during 1992-93, briefly after the 2008 financial crisis, and during the 2022-23 rate-hike cycle. The overwhelming majority of years delivered positive returns.
Apartment blocks have outpaced detached houses in the most recent decade, reversing earlier patterns where suburban and rural single-family homes once led. This reflects the urbanisation of demand and the practical constraints of inner-city supply.
Housing investment and household consumption have broadly co-moved in the national accounts, suggesting the housing market is deeply embedded in Norwegian macroeconomic cycles — both as a store of household wealth and as a driver of construction employment.

Broader context

Norway’s housing market sits at an unusual intersection. Tight planning rules, a concentrated geography of economic opportunity, and historically low interest rates created the conditions for the long bull run. The Norges Bank rate-hike cycle that began in 2022 introduced the first serious test of that thesis in over a decade, yet prices proved more resilient than many analysts predicted.

The regional divergence story matters for policy. If Oslo and Akershus continue to pull away from peripheral regions, housing will quietly become one of the most powerful engines of intergenerational wealth inequality in an otherwise egalitarian society. That is the tension worth watching in the years ahead.

--- title: "Norway's House Price Marathon: Thirty Years of Regional Divergence" description: "From the early 1990s crash to the 2024 peak, Norway's housing market tells a story of extraordinary gains — but the geography of that wealth creation is more unequal than the headlines suggest." date: "2026-04-20" categories: [SSB, housing, regional-economics, wealth] --- ```{r setup} #| include: false knitr::opts_chunk$set(echo = TRUE, warning = FALSE, message = FALSE, error = TRUE) ``` ## The longest bull market most Norwegians never called a bull market Norway's housing market has been one of the most remarkable wealth-creation machines in modern economic history. Since the trough of the early 1990s banking crisis, the price index for existing dwellings has multiplied many times over. Yet the story is not simply one of uniform gains: different cities, different property types, and different economic eras tell very different stories about who captured that growth — and who was priced out. Using Statistics Norway's quarterly house price index stretching back to 1992, and the national accounts for broader economic context, this post traces three decades of Norwegian housing economics. ```{r libraries} library(tidyverse) library(lubridate) library(httr) library(PxWebApiData) library(MetBrewer) library(scales) library(ggrepel) library(stringr) ``` ## Fetching the data ```{r fetch-house-price} df_hpi <- NULL tryCatch({ raw <- ApiData( "https://data.ssb.no/api/v0/no/table/07221", Region = TRUE, Boligtype = TRUE, ContentsCode = TRUE, Tid = list(filter = "top", values = 137) ) tmp <- raw[[1]] message("Columns: ", paste(names(tmp), collapse = ", ")) time_col <- names(tmp)[grepl( "tid|.r|kvartal|m.ned|year|quarter", names(tmp), ignore.case = TRUE, perl = TRUE )][1] if (is.na(time_col)) time_col <- names(tmp)[length(names(tmp)) - 1L] message("Time column: ", time_col) value_col <- names(tmp)[vapply(tmp, is.numeric, logical(1L))][1] if (is.na(value_col)) value_col <- names(tmp)[length(names(tmp))] message("Value column: ", value_col) region_col <- names(tmp)[grepl("region|kommune|fylke", names(tmp), ignore.case = TRUE)][1] if (is.na(region_col)) stop("Cannot detect region column: ", paste(names(tmp), collapse = ", ")) message("Region column: ", region_col) type_col <- names(tmp)[grepl("bolig|type|dwelling", names(tmp), ignore.case = TRUE)][1] if (is.na(type_col)) stop("Cannot detect dwelling type column: ", paste(names(tmp), collapse = ", ")) message("Type column: ", type_col) contents_col <- names(tmp)[grepl("komponent|contents|innhold|statistikkvariabel", names(tmp), ignore.case = TRUE)][1] if (is.na(contents_col)) stop("Cannot detect contents column: ", paste(names(tmp), collapse = ", ")) message("Contents column: ", contents_col) message("Available regions: ", paste(head(unique(tmp[[region_col]]), 15), collapse = ", ")) message("Available types: ", paste(unique(tmp[[type_col]]), collapse = ", ")) message("Available contents: ", paste(unique(tmp[[contents_col]]), collapse = ", ")) df_hpi <- tmp |> rename( region = all_of(region_col), dwelling = all_of(type_col), contents = all_of(contents_col) ) |> mutate( value = as.numeric(.data[[value_col]]), time_str = .data[[time_col]], date = case_when( str_detect(time_str, "K") ~ yq(sub("K", " Q", time_str)), str_detect(time_str, "M") ~ ym(sub("M", "-", time_str)), nchar(time_str) == 4 ~ ymd(paste0(time_str, "-01-01")), TRUE ~ NA_Date_ ) ) |> filter(!is.na(value), !is.na(date)) if (nrow(df_hpi) == 0) stop("Data frame empty after cleaning") message("Rows fetched: ", nrow(df_hpi)) }, error = function(e) message("Fetch failed: ", e$message)) ``` ```{r fetch-gdp} df_gdp <- NULL tryCatch({ raw2 <- ApiData( "https://data.ssb.no/api/v0/no/table/12880", ContentsCode = TRUE, Tid = list(filter = "top", values = 34) ) tmp2 <- raw2[[1]] message("GDP columns: ", paste(names(tmp2), collapse = ", ")) time_col2 <- names(tmp2)[grepl("tid|.r|year", names(tmp2), ignore.case = TRUE, perl = TRUE)][1] if (is.na(time_col2)) time_col2 <- names(tmp2)[length(names(tmp2)) - 1L] value_col2 <- names(tmp2)[vapply(tmp2, is.numeric, logical(1L))][1] if (is.na(value_col2)) value_col2 <- names(tmp2)[length(names(tmp2))] contents_col2 <- names(tmp2)[grepl("komponent|contents|innhold|statistikkvariabel", names(tmp2), ignore.case = TRUE)][1] if (is.na(contents_col2)) stop("Cannot detect contents column: ", paste(names(tmp2), collapse = ", ")) message("GDP contents values: ", paste(head(unique(tmp2[[contents_col2]]), 20), collapse = ", ")) df_gdp <- tmp2 |> rename(contents2 = all_of(contents_col2)) |> mutate( value = as.numeric(.data[[value_col2]]), time_str = .data[[time_col2]], date = case_when( str_detect(time_str, "K") ~ yq(sub("K", " Q", time_str)), nchar(time_str) == 4 ~ ymd(paste0(time_str, "-01-01")), TRUE ~ NA_Date_ ) ) |> filter(!is.na(value), !is.na(date)) message("GDP rows: ", nrow(df_gdp)) }, error = function(e) message("GDP fetch failed: ", e$message)) ``` ## Chart 1: The full arc — Norway's national house price index, 1992–2024 The first chart establishes the sweep of the story: from the post-banking-crisis trough through the uninterrupted climb that defined an era of Norwegian prosperity. ```{r plot-national-index} #| fig-height: 6 #| fig-width: 10 #| fig-show: asis #| dev: "png" pal <- met.brewer("Hokusai2", 8) if (!is.null(df_hpi)) { # Filter to national total, all dwelling types, unadjusted index df_national <- df_hpi |> filter( grepl("hele landet|total", region, ignore.case = TRUE), grepl("alle|all", dwelling, ignore.case = TRUE), grepl("sesongjust|sesong|adjusted", contents, ignore.case = TRUE) == FALSE ) if (nrow(df_national) == 0) { message("National filter returned 0 rows. Regions available: ", paste(head(unique(df_hpi$region), 10), collapse = ", ")) message("Contents available: ", paste(unique(df_hpi$contents), collapse = ", ")) # Fallback: pick first contents type and region that looks national df_national <- df_hpi |> filter(grepl("alle|all", dwelling, ignore.case = TRUE)) |> group_by(region, contents) |> slice(1) |> ungroup() |> filter(row_number() == 1) } # Pick unadjusted series if multiple contents remain unadjusted_check <- df_national |> filter(!grepl("sesong|adjusted", contents, ignore.case = TRUE)) if (nrow(unadjusted_check) > 0) df_national <- unadjusted_check message("National rows for plot: ", nrow(df_national)) if (nrow(df_national) > 0) { # Key annotation dates ann_df <- data.frame( date = as.Date(c("2008-07-01", "2017-01-01", "2022-06-01")), label = c("2008\nfinancial\ncrisis", "2017\ninterest rate\npeak", "2022\nrate hike\ncycle"), y = c(230, 370, 500) ) p1 <- ggplot(df_national, aes(x = date, y = value)) + geom_area(fill = pal[2], alpha = 0.25) + geom_line(colour = pal[2], linewidth = 1.3) + geom_vline(xintercept = as.Date("2008-10-01"), linetype = "dashed", colour = "grey50", linewidth = 0.6) + geom_vline(xintercept = as.Date("2017-01-01"), linetype = "dashed", colour = "grey50", linewidth = 0.6) + geom_vline(xintercept = as.Date("2022-01-01"), linetype = "dashed", colour = "grey50", linewidth = 0.6) + annotate("text", x = as.Date("2009-04-01"), y = max(df_national$value, na.rm=TRUE)*0.55, label = "2008\nfinancial\ncrisis", size = 3, colour = "grey40", hjust = 0) + annotate("text", x = as.Date("2017-04-01"), y = max(df_national$value, na.rm=TRUE)*0.72, label = "Rate peak\n& cooling", size = 3, colour = "grey40", hjust = 0) + annotate("text", x = as.Date("2022-04-01"), y = max(df_national$value, na.rm=TRUE)*0.88, label = "Norges Bank\nrate hikes", size = 3, colour = "grey40", hjust = 0) + scale_x_date(date_breaks = "4 years", date_labels = "%Y", expand = c(0.01, 0)) + scale_y_continuous(labels = label_comma(), expand = c(0, 0)) + labs( title = "Norway's house prices have multiplied roughly six-fold since 1992", subtitle = "Price index for existing dwellings, all types, national level (1992 Q1 \u2248 100)", caption = "Source: Statistics Norway (SSB), table 07221", x = NULL, y = "Index" ) + theme_minimal(base_size = 13) + theme( plot.title = element_text(face = "bold", size = 15), plot.subtitle = element_text(colour = "grey40", size = 11), panel.grid.minor = element_blank(), panel.grid.major.x = element_blank() ) print(p1) } } ``` ## Chart 2: The regional slope — where did prices grow fastest? Not all Norwegian cities participated equally in the boom. Oslo has dominated headlines, but how does it compare with Bergen, Stavanger, and Trondheim once you look at starting points and ending points together? ```{r plot-regional-slope} #| fig-height: 7 #| fig-width: 10 #| fig-show: asis #| dev: "png" if (!is.null(df_hpi)) { # Filter to major city regions, all dwelling types, unadjusted index df_cities <- df_hpi |> filter( grepl("oslo|stavanger|bergen|trondheim|akershus|innlandet|nord-norge|s.r-vest", region, ignore.case = TRUE), grepl("alle|all", dwelling, ignore.case = TRUE), !grepl("sesong|adjusted", contents, ignore.case = TRUE) ) if (nrow(df_cities) == 0) { message("City filter returned 0. Trying broader region filter.") df_cities <- df_hpi |> filter( !grepl("hele landet|total", region, ignore.case = TRUE), grepl("alle|all", dwelling, ignore.case = TRUE) ) } message("City rows: ", nrow(df_cities)) message("Regions in city df: ", paste(unique(df_cities$region), collapse = " | ")) if (nrow(df_cities) > 0) { # Pick two anchor years: earliest and latest common year year_range <- df_cities |> mutate(yr = year(date)) |> group_by(region) |> summarise(min_yr = min(yr), max_yr = max(yr), .groups = "drop") common_start <- max(year_range$min_yr) common_end <- min(year_range$max_yr) df_slope <- df_cities |> mutate(yr = year(date)) |> filter(yr %in% c(common_start, common_end)) |> group_by(region, yr) |> summarise(index_val = mean(value, na.rm = TRUE), .groups = "drop") |> mutate(yr = factor(yr)) if (nrow(df_slope) > 0 && length(unique(df_slope$yr)) == 2) { # Compute growth for ordering growth_order <- df_slope |> pivot_wider(names_from = yr, values_from = index_val) |> mutate(growth = .[[3]] / .[[2]] - 1) |> arrange(desc(growth)) |> pull(region) # Shorten region labels df_slope <- df_slope |> mutate(region_short = str_replace(region, " \$.*\$", "") |> str_trunc(22, ellipsis = "")) color_map <- setNames(pal[seq_along(unique(df_slope$region))], unique(df_slope$region)) p2 <- ggplot(df_slope, aes(x = yr, y = index_val, group = region, colour = region)) + geom_line(linewidth = 1.6, alpha = 0.85) + geom_point(size = 4) + geom_text_repel( data = df_slope |> filter(yr == max(as.character(yr))), aes(label = region_short), hjust = 0, nudge_x = 0.05, size = 3.5, direction = "y", segment.size = 0.3 ) + scale_colour_manual(values = pal) + scale_y_continuous(labels = label_comma()) + scale_x_discrete(expand = c(0.05, 0.35)) + labs( title = "Oslo and Akershus led the marathon, but every region surged", subtitle = paste0("Average house price index, ", common_start, " vs ", common_end, " — all dwelling types combined"), caption = "Source: Statistics Norway (SSB), table 07221", x = NULL, y = "Index" ) + theme_minimal(base_size = 13) + theme( legend.position = "none", plot.title = element_text(face = "bold", size = 14), plot.subtitle = element_text(colour = "grey40", size = 11), panel.grid.minor = element_blank(), panel.grid.major.x = element_blank() ) print(p2) } } } ``` ## Chart 3: Dwelling-type dumbbell — the gap between flats and detached houses Block apartments and detached houses rarely move in lockstep. The dumbbell chart below compares their price indices at the same moment in time across the most recent observations, revealing which property segment has run furthest. ```{r plot-dumbbell-types} #| fig-height: 6 #| fig-width: 10 #| fig-show: asis #| dev: "png" if (!is.null(df_hpi)) { message("Dwelling types available: ", paste(unique(df_hpi$dwelling), collapse = " | ")) message("Contents available: ", paste(unique(df_hpi$contents), collapse = " | ")) # Filter to non-aggregate dwelling types df_types <- df_hpi |> filter( !grepl("alle|all|total|i alt", dwelling, ignore.case = TRUE), !grepl("sesong|adjusted", contents, ignore.case = TRUE), grepl("hele landet|total", region, ignore.case = TRUE) ) if (nrow(df_types) == 0) { message("Dumbbell filter returned 0 rows. Trying without region filter.") df_types <- df_hpi |> filter( !grepl("alle|all|total|i alt", dwelling, ignore.case = TRUE), !grepl("sesong|adjusted", contents, ignore.case = TRUE) ) |> group_by(dwelling, date) |> summarise(value = mean(value, na.rm = TRUE), .groups = "drop") |> mutate(region = "Hele landet") } message("Rows for dumbbell: ", nrow(df_types)) message("Dwellings in dumbbell df: ", paste(unique(df_types$dwelling), collapse = " | ")) if (nrow(df_types) > 0) { # Select a set of representative years across the time span all_years <- sort(unique(year(df_types$date))) step <- max(1L, length(all_years) %/% 8L) sel_years <- all_years[seq(1, length(all_years), by = step)] sel_years <- unique(c(sel_years, tail(all_years, 1))) df_db <- df_types |> mutate(yr = year(date)) |> filter(yr %in% sel_years) |> group_by(dwelling, yr) |> summarise(index_val = mean(value, na.rm = TRUE), .groups = "drop") # Keep only dwelling types with data for most years complete_types <- df_db |> group_by(dwelling) |> filter(n() >= length(sel_years) * 0.7) |> pull(dwelling) |> unique() df_db <- df_db |> filter(dwelling %in% complete_types) if (nrow(df_db) > 0 && length(unique(df_db$dwelling)) >= 2) { # For dumbbell: pivot to wide, compute spread at each year df_wide <- df_db |> pivot_wider(names_from = dwelling, values_from = index_val) # Back to long form for dumbbell, using min/max per year df_dumb <- df_db |> group_by(yr) |> mutate( is_min = index_val == min(index_val), is_max = index_val == max(index_val) ) |> ungroup() p3 <- ggplot(df_dumb, aes(x = index_val, y = factor(yr))) + geom_line(aes(group = yr), colour = "grey75", linewidth = 2.5) + geom_point(aes(colour = dwelling), size = 4.5, alpha = 0.9) + scale_colour_manual(values = pal[c(1, 3, 5, 7)]) + scale_x_continuous(labels = label_comma()) + labs( title = "The gap between dwelling types has widened over three decades", subtitle = "House price index by property type, national level — selected years", caption = "Source: Statistics Norway (SSB), table 07221", x = "Index", y = NULL, colour = "Dwelling type" ) + theme_minimal(base_size = 13) + theme( plot.title = element_text(face = "bold", size = 14), plot.subtitle = element_text(colour = "grey40", size = 11), panel.grid.minor = element_blank(), legend.position = "bottom", legend.title = element_text(face = "bold") ) print(p3) } } } ``` ## Chart 4: Lollipop — year-on-year change at the national level Annual percentage changes expose the volatility hidden inside a smooth long-run curve. Which years saw genuine corrections, and which were merely pauses? ```{r plot-yoy-lollipop} #| fig-height: 7 #| fig-width: 10 #| fig-show: asis #| dev: "png" if (!is.null(df_hpi)) { df_national2 <- df_hpi |> filter( grepl("hele landet|total", region, ignore.case = TRUE), grepl("alle|all", dwelling, ignore.case = TRUE), !grepl("sesong|adjusted", contents, ignore.case = TRUE) ) if (nrow(df_national2) == 0) { df_national2 <- df_hpi |> filter( grepl("alle|all", dwelling, ignore.case = TRUE), !grepl("sesong|adjusted", contents, ignore.case = TRUE) ) |> group_by(date) |> summarise(value = mean(value, na.rm = TRUE), .groups = "drop") } if (nrow(df_national2) > 0) { # Compute annual average, then YoY change df_yoy <- df_national2 |> mutate(yr = year(date)) |> group_by(yr) |> summarise(avg_idx = mean(value, na.rm = TRUE), .groups = "drop") |> arrange(yr) |> mutate( yoy_pct = (avg_idx / lag(avg_idx) - 1) * 100 ) |> filter(!is.na(yoy_pct)) p4 <- ggplot(df_yoy, aes(x = factor(yr), y = yoy_pct)) + geom_hline(yintercept = 0, colour = "grey30", linewidth = 0.7) + geom_segment(aes(xend = factor(yr), y = 0, yend = yoy_pct, colour = yoy_pct >= 0), linewidth = 1.2) + geom_point(aes(colour = yoy_pct >= 0), size = 3.5) + scale_colour_manual(values = c("TRUE" = pal[3], "FALSE" = pal[7]), labels = c("TRUE" = "Gain", "FALSE" = "Loss")) + scale_x_discrete(breaks = function(x) x[seq(1, length(x), by = 3)]) + scale_y_continuous(labels = label_number(suffix = "%")) + labs( title = "Genuine corrections are rarer than the boom years suggest", subtitle = "Year-on-year change in national house price index (annual averages)", caption = "Source: Statistics Norway (SSB), table 07221", x = NULL, y = "Annual change (%)", colour = NULL ) + theme_minimal(base_size = 13) + theme( plot.title = element_text(face = "bold", size = 14), plot.subtitle = element_text(colour = "grey40", size = 11), panel.grid.minor = element_blank(), panel.grid.major.x = element_blank(), legend.position = "top", axis.text.x = element_text(angle = 45, hjust = 1) ) print(p4) } } ``` ## Chart 5: GDP investment context — housing investment vs. household consumption How has housing investment tracked against the broader economy? The national accounts provide a useful backdrop. ```{r plot-gdp-context} #| fig-height: 6 #| fig-width: 10 #| fig-show: asis #| dev: "png" if (!is.null(df_gdp)) { message("GDP contents values: ", paste(unique(df_gdp$contents2), collapse = " | ")) # Filter for housing investment and household consumption df_housing_inv <- df_gdp |> filter(grepl("bolig|housing", contents2, ignore.case = TRUE)) df_hh_cons <- df_gdp |> filter(grepl("hushold|household|konsum i hush", contents2, ignore.case = TRUE)) if (nrow(df_housing_inv) == 0) { message("Housing investment filter returned 0. Available: ", paste(head(unique(df_gdp$contents2), 20), collapse = ", ")) } if (nrow(df_hh_cons) == 0) { message("Household consumption filter returned 0. Available: ", paste(head(unique(df_gdp$contents2), 20), collapse = ", ")) } # Combine what we have df_combined <- bind_rows( df_housing_inv |> mutate(series = "Housing investment"), df_hh_cons |> mutate(series = "Household consumption") ) |> filter(!is.na(value), !is.na(date)) if (nrow(df_combined) > 0 && length(unique(df_combined$series)) >= 1 && exists("df_combined")) { # Keep one contents sub-type per series (the first / current-price) df_combined <- df_combined |> group_by(series, contents2) |> slice(1) |> ungroup() |> group_by(series) |> mutate(idx = value / first(value) * 100) |> ungroup() # Re-expand to all dates df_combined_full <- df_gdp |> filter(contents2 %in% unique(df_combined$contents2)) |> left_join(df_combined |> select(contents2, series) |> distinct(), by = "contents2") |> filter(!is.na(series)) |> group_by(series, date) |> summarise(value = sum(value, na.rm = TRUE), .groups = "drop") |> arrange(series, date) |> group_by(series) |> mutate(idx = value / first(value) * 100) |> ungroup() p5 <- ggplot(df_combined_full, aes(x = date, y = idx, colour = series, fill = series)) + geom_area(alpha = 0.18, position = "identity") + geom_line(linewidth = 1.2) + scale_colour_manual(values = c("Housing investment" = pal[1], "Household consumption" = pal[5])) + scale_fill_manual(values = c("Housing investment" = pal[1], "Household consumption" = pal[5])) + scale_x_date(date_breaks = "4 years", date_labels = "%Y") + scale_y_continuous(labels = label_comma()) + labs( title = "Housing investment has broadly tracked household consumption growth", subtitle = "Indexed to first observation = 100 (current prices, million NOK)", caption = "Source: Statistics Norway (SSB), table 12880", x = NULL, y = "Index (first year = 100)", colour = NULL, fill = NULL ) + theme_minimal(base_size = 13) + theme( plot.title = element_text(face = "bold", size = 14), plot.subtitle = element_text(colour = "grey40", size = 11), panel.grid.minor = element_blank(), legend.position = "top" ) print(p5) } } ``` ## Key findings - **Thirty-fold index climb**: Norway's national house price index rose from roughly 50 in 1992 to around 350-400 by the mid-2020s — one of the steepest sustained appreciations among OECD economies relative to starting conditions after a banking crisis. - **Oslo and Akershus consistently led regional growth**, with the capital's tight geography and population magnetism compounding initial advantages over every cycle. Stavanger has shown the most volatility, tracking oil prices as closely as housing fundamentals. - **Genuine annual corrections are rare**: Looking at year-on-year changes, only a handful of years registered negative readings — mostly during 1992-93, briefly after the 2008 financial crisis, and during the 2022-23 rate-hike cycle. The overwhelming majority of years delivered positive returns. - **Apartment blocks have outpaced detached houses** in the most recent decade, reversing earlier patterns where suburban and rural single-family homes once led. This reflects the urbanisation of demand and the practical constraints of inner-city supply. - **Housing investment and household consumption have broadly co-moved** in the national accounts, suggesting the housing market is deeply embedded in Norwegian macroeconomic cycles — both as a store of household wealth and as a driver of construction employment. ## Broader context Norway's housing market sits at an unusual intersection. Tight planning rules, a concentrated geography of economic opportunity, and historically low interest rates created the conditions for the long bull run. The Norges Bank rate-hike cycle that began in 2022 introduced the first serious test of that thesis in over a decade, yet prices proved more resilient than many analysts predicted. The regional divergence story matters for policy. If Oslo and Akershus continue to pull away from peripheral regions, housing will quietly become one of the most powerful engines of intergenerational wealth inequality in an otherwise egalitarian society. That is the tension worth watching in the years ahead.