"""New homes per resident in New York, Tokyo, London, the Paris region and Toronto: the analysis behind a SparkyData article. py src/build_article.py homes-per-resident How fast New York adds homes for its population against four other big cities, now and over about thirty years. Each city's own measure is kept and named (cards ``census_bps``, ``housing_production``, ``world_city_housing``; populations from ``world_city_population``): - New York: homes permitted (the Census Bureau's Building Permits Survey, the five boroughs, 1990 on) and homes completed (the Department of City Planning's Housing Database, net of demolitions, 2010 on); - Tokyo (Tokyo-to): housing starts from building-works notices (MLIT, 1951 on), the measure closest to a permit; - London: net additional dwellings (MHCLG Table 118, financial years from 2000-01), the official measure of supply, completions less demolitions; the building-control starts undercount and are shown apart; - the Paris region (Île-de-France): dwellings started and authorised (SDES, 2002 on); - Toronto (the census metropolitan area): starts and completions (CMHC, 1972 on; population from 2001). A rate is homes per 1,000 residents in the same year. The ten-year average (2016-2025, London's ten financial years 2015-16 to 2024-25) is the comparison least moved by one year's rush; the current year is compared over the same months with last year (New York and Tokyo January to September; Toronto and the Paris region January to September). A change between two counts of homes is not a count of independent events, so no chance range is drawn for it; it is reported as the change it is. Writes results.json, viz.json, charts.json, data.csv and build.json into output/articles/homes_per_resident/ and site/public/articles/homes-per-resident/, and copies this script there as analysis.py. Deterministic: no sampling, and SEED is kept for any resampling. """ from __future__ import annotations import re import pandas as pd from article_kit import Article, lines, r, strip from nycdata import housing_db, permits, world_housing as H, world_population as P SLUG = "homes-per-resident" SEED = 20261104 FIRST, LAST = 1990, 2025 DECADE = (2016, 2025) BOROUGHS = ["Bronx", "Brooklyn", "Manhattan", "Queens", "Staten Island"] COY = ("dcp_city_of_yes_housing_final_plan", "dcp_city_of_yes_housing_opportunity_final_plan.pdf") SDES_NOTE = ("sdes_construction_logements_release_2026_09", "sdes_construction_logements_2026_09.html") MONTH_NAMES = ["January", "February", "March", "April", "May", "June", "July", "August", "September", "October", "November", "December"] #: the measure each city is compared on, and its label MAIN = [("nyc_permits", "New York, homes permitted"), ("tokyo", "Tokyo, homes started"), ("toronto", "Toronto area, homes started"), ("paris", "Paris region, homes started"), ("london", "London, net homes added")] def homes_by_measure() -> dict[str, pd.Series]: """Every yearly count, indexed by calendar year (London's financial year by the year it starts in).""" y = permits.yearly() comp = housing_db.completions().set_index("year") pm = H.paris_monthly() idf = pm[pm["area"].eq("ile_de_france")] tm = H.toronto_monthly() lq = H.london_quarterly() return {"nyc_permits": y[y["area"].isin(BOROUGHS)].groupby("year")["units"].sum(), "nyc_completions": comp["net"], "tokyo": H.tokyo_annual().set_index("year")["starts"], "toronto": H.annual(tm, "starts"), "toronto_completions": H.annual(tm, "completions"), "paris": H.annual(idf, "started"), "paris_authorised": H.annual(idf, "authorised"), "london": H.london_net().set_index("fy")["net"], "london_bc_starts": H.annual(lq, "starts"), "london_bc_completions": H.annual(lq, "completions")} AREA = {"nyc_permits": "new_york", "nyc_completions": "new_york", "tokyo": "tokyo", "toronto": "toronto_cma", "toronto_completions": "toronto_cma", "paris": "ile_de_france", "paris_authorised": "ile_de_france", "london": "london", "london_bc_starts": "london", "london_bc_completions": "london"} def main(argv: list[str] | None = None) -> int: """Run the analysis and write every file (no options).""" A = Article(SLUG, __file__, seed=SEED) s = homes_by_measure() per = {k: P.per_1000(v.astype(float), AREA[k]) for k, v in s.items()} # ---- the decade, every measure lo, hi = DECADE dec = {} for k, v in per.items(): w = v.loc[lo - 1:hi - 1] if k == "london" else v.loc[lo:hi] need = 10 dec[k] = {"per_1000": r(w.mean(), 2), "years": int(w.notna().sum()), "first": int(w.index.min()), "last": int(w.index.max())} assert dec[k]["years"] == need, (k, dec[k]) A.results["decade"] = dec order = sorted((k for k, _ in MAIN), key=lambda k: dec[k]["per_1000"]) assert order[0] == "nyc_permits", order # the page: New York's is the lowest # New York's completions, its other measure, are below every other city's main measure too assert all(dec["nyc_completions"]["per_1000"] < dec[k]["per_1000"] for k, _ in MAIN if k != "nyc_permits") A.results["decade_times"] = {"tokyo_vs_nyc": r(dec["tokyo"]["per_1000"] / dec["nyc_permits"]["per_1000"], 1)} ln = H.london_net().set_index("fy") ex = P.per_1000((ln["net"] - ln["census_adjustment"]).astype(float), "london").loc[lo - 1:hi - 1] A.results["london_ex_adjustment"] = r(ex.mean(), 2) pm_ = H.paris_monthly() core = {"tokyo_wards": P.per_1000(H.tokyo_wards_annual().set_index("year")["starts"].astype(float), "tokyo_wards"), "toronto_city": P.per_1000(H.toronto_city_annual().set_index("year")["starts"].astype(float), "toronto"), "paris_city": P.per_1000(H.annual(pm_[pm_["area"].eq("paris")], "started").astype(float), "paris"), "paris_inner": P.per_1000(H.annual(pm_[pm_["area"].eq("paris_inner")], "started").astype(float), "paris_inner")} A.results["core_decade"] = {k: r(v.loc[lo:hi].mean(), 1) for k, v in core.items()} assert all(v.loc[lo:hi].notna().sum() == 10 for v in core.values()) early = {k: (per[k].loc[2005:2014] if k == "london" else per[k].loc[2006:2015]) for k, _ in MAIN} A.results["early_decade"] = {k: r(v.mean(), 2) for k, v in early.items()} assert min(A.results["early_decade"], key=A.results["early_decade"].get) == "nyc_permits" # ---- the long view: Tokyo above New York in every year since 1990 years = list(range(FIRST, LAST + 1)) ny, tk = per["nyc_permits"], per["tokyo"] assert all(tk[t] > ny[t] for t in years), "the page says Tokyo was above New York every year" tokyo_pop = P.population("tokyo").set_index("year")["pop"] revised = tokyo_pop.where(tokyo_pop.index < 2021, tokyo_pop * 0.995) # half a percent lower from 2021 (assumed) tk_rev = 1000 * s["tokyo"].astype(float).reindex(revised.index).dropna() / revised tokyo_sensitivity = {"decade": r(tk_rev.loc[2016:2025].mean(), 1), "y2025": r(tk_rev[LAST], 1)} assert tokyo_sensitivity == {"decade": r(per["tokyo"].loc[2016:2025].mean(), 1), "y2025": r(tk[LAST], 1)}, tokyo_sensitivity A.results["tokyo_revision_sensitivity"] = { "population_lower_pct": 0.5, "from_year": 2021, "decade": r(tk_rev.loc[2016:2025].mean(), 2), "y2025": r(tk_rev[LAST], 2), "decade_as_read": r(per["tokyo"].loc[2016:2025].mean(), 2), "y2025_as_read": r(tk[LAST], 2), "assumption": True, "reason": ("TMG revised November 2020 to September 2026 to the 2025 census on 29 October 2026, lowering " "1 January 2026 from 14,270,748 to 14,234,309 (0.26%; source tmg_population_monthly's note); the " "revised table was not at a fetchable address, so 2021-2025 are lowered by 0.5%, about twice that " "share, and the printed Tokyo figures are checked unchanged at one decimal")} ny_peak = int(ny.loc[FIRST:LAST].idxmax()) A.results["long"] = { "tokyo_above_every_year": True, "nyc_1990s_max": r(ny.loc[1990:1999].max(), 1), "nyc_peak_year": ny_peak, "nyc_peak": r(ny[ny_peak], 1), "nyc_peak_units": int(s["nyc_permits"][ny_peak]), "tokyo_peak_year_rate": r(tk[ny_peak], 1), "nyc_2005_2008_min": r(ny.loc[2005:2008].min(), 1), "nyc_2005_2008_max": r(ny.loc[2005:2008].max(), 1), "paris_2005_2008_min": r(per["paris"].loc[2005:2008].min(), 1), "paris_2005_2008_max": r(per["paris"].loc[2005:2008].max(), 1), "nyc_2009": r(ny[2009], 1), "tokyo_1990": r(tk[1990], 1), "tokyo_2003": r(tk[2003], 1), "tokyo_2025": r(tk[LAST], 1), "tokyo_2025_units": int(s["tokyo"][LAST]), "nyc_2025": r(ny[LAST], 1), "nyc_2025_units": int(s["nyc_permits"][LAST]), "tokyo_vs_nyc_2025": r(tk[LAST] / ny[LAST], 1), } A.results["long"]["nyc_back_at_2005_pace"] = [int(t) for t in range(2009, LAST + 1) if ny[t] >= ny.loc[2005:2008].min()] assert A.results["long"]["nyc_back_at_2005_pace"] == [2015, 2022], A.results["long"]["nyc_back_at_2005_pace"] # ---- 2025, each city's measures A.results["y2025"] = {k: {"count": int(s[k][LAST]) if LAST in s[k].index else None, "per_1000": r(per[k][LAST], 2) if LAST in per[k].index else None} for k in s} A.results["y2025"]["london"] = {"count": int(s["london"][2024]), "per_1000": r(per["london"][2024], 2), "label": "2024-25"} for k in ("london_bc_starts", "london_bc_completions"): # calendar years, unlike the net count's April-to-March A.results["y2025"][k]["period"] = "calendar 2025 (four quarters)" comp = housing_db.completions().set_index("year") A.results["nyc_completions_2025"] = {"net": int(comp.loc[LAST, "net"]), "new_build": int(comp.loc[LAST, "new_build"]), "first_year": int(comp.index.min()), "is_highest": bool(comp["net"].idxmax() == LAST)} assert A.results["nyc_completions_2025"]["is_highest"] one = {"tokyo": per["tokyo"][LAST], "toronto": per["toronto"][LAST], "paris": per["paris"][LAST], "london": per["london"][2024]} A.results["y2025_main"] = {k: r(v, 1) for k, v in one.items()} # the page: New York's 2025 permits sat below all four other main measures; its net completions above three A.results["y2025_nyc_permits_lowest"] = bool(all(A.results["y2025"]["nyc_permits"]["per_1000"] < v for v in one.values())) assert A.results["y2025_nyc_permits_lowest"] assert all(per["nyc_completions"][LAST] > v for k, v in one.items() if k != "tokyo") and per["nyc_completions"][LAST] < one["tokyo"] py = housing_db.permit_years(LAST) A.results["nyc_completions_2025"]["permitted_2022_pct"] = r(100 * py.get(2022, 0.0), 0) A.results["nyc_completions_2025"]["known_permit_year_pct"] = r(100 * housing_db.permit_year_coverage(LAST), 1) mt_ = H.tokyo_monthly() m25 = mt_[mt_["year"] == LAST].set_index("month")["starts"] A.results["tokyo_march_2025"] = {"starts": int(m25[3]), "median_other_months": int(m25.drop(3).median())} # ---- this year, the same months as last year tm, mt = H.toronto_monthly(), H.tokyo_monthly() pm = H.paris_monthly() idf = pm[pm["area"].eq("ile_de_france")] p26, p25 = permits.so_far(2026, 9), permits.so_far(2025, 9) now = { "nyc": {"months": 9, "now": int(p26[p26["area"].isin(BOROUGHS)]["units"].sum()), "before": int(p25[p25["area"].isin(BOROUGHS)]["units"].sum()), "area": "new_york"}, "tokyo": {"months": 9, "now": int(H.so_far(mt, "starts", 9)[2026]), "before": int(H.so_far(mt, "starts", 9)[2025]), "area": "tokyo"}, "toronto": {"months": 9, "now": int(H.so_far(tm, "starts", 9)[2026]), "before": int(H.so_far(tm, "starts", 9)[2025]), "area": "toronto_cma"}, "paris": {"months": 9, "now": int(H.so_far(idf, "started", 9)[2026]), "before": int(H.so_far(idf, "started", 9)[2025]), "area": "ile_de_france"}, } for k, v in now.items(): pop = P.population(v["area"]).set_index("year")["pop"] v["pop_year"] = int(2026 if 2026 in pop.index else pop.index.max()) # 2025 except the Paris region's 1 January 2026 p = pop[v["pop_year"]] v["change_pct"] = r(100 * (v["now"] / v["before"] - 1), 0) v["per_1000"] = r(1000 * v["now"] / p, 2) lq = H.london_quarterly().set_index(["year", "quarter"])["starts"] now["london_bc"] = {"months": 6, "now": int(lq.loc[(2026, 1)] + lq.loc[(2026, 2)]), "before": int(lq.loc[(2025, 1)] + lq.loc[(2025, 2)])} A.results["now"] = now lq4 = H.london_quarterly().set_index(["year", "quarter"])["completions"] ln = H.london_net().set_index("fy") A.results["london_undercount"] = {"label": "2024-25", "period": "April 2024 to March 2025, both counts", "new_build": int(round(ln.loc[2024, "new_build"], -1)), "bc_completions": int(lq4.loc[(2024, 2)] + lq4.loc[(2024, 3)] + lq4.loc[(2024, 4)] + lq4.loc[(2025, 1)])} A.results["now_times"] = {"tokyo_vs_nyc": r(now["tokyo"]["per_1000"] / now["nyc"]["per_1000"], 1), "nyc_lead_over_toronto_pct": r(100 * (now["nyc"]["per_1000"] / now["toronto"]["per_1000"] - 1), 0)} assert now["nyc"]["per_1000"] > now["toronto"]["per_1000"] and now["nyc"]["per_1000"] < now["paris"]["per_1000"] A.latest("2026-09", "Census Bureau permits, MLIT, CMHC and SDES starts, January to September 2026 (the same months as 2025)") A.long_view(FIRST, LAST, "New York's homes permitted and Tokyo's homes started per 1,000 residents, 1990-2025 " "(Toronto from 2001, the Paris region from 2002, London from 2000-01)") # ---- the documents Q = A.quotes() quotes = {} for key, needle in {"coy_adopted": "On December 5, the City Council approved City of Yes for Housing Opportunity with modifications", "coy_82000": "Enables the creation of 82,000 new homes over the next 15 years"}.items(): q = Q.quote(COY[0], COY[1], needle) quotes[key] = {"quote": q["quote"], "page": q.get("page"), "source": q.get("source")} for key, needle in {"sitadel3": "À compter de mars 2026, les données sur la construction neuve de logements sont produites " "à partir du dispositif Sitadel3, qui remplace Sitadel2"}.items(): q = Q.quote(SDES_NOTE[0], SDES_NOTE[1], needle) quotes[key] = {"quote": q["quote"], "page": q.get("page"), "source": q.get("source")} A.results["quotes"] = quotes A.results["history"] = {"coy_homes": int(re.search(r"([\d,]+) new homes", quotes["coy_82000"]["quote"]).group(1).replace(",", "")), "coy_years": int(re.search(r"next (\d+) years", quotes["coy_82000"]["quote"]).group(1))} # ---- the reproduction: the figures the agencies published A.reproduce("Tokyo homes started, 2025 (MLIT annual table)", published=122130, published_moe90=None, reproduced=int(s["tokyo"][LAST]), kind="count", tolerance=0) A.reproduce("Tokyo homes started, 2025 (the twelve monthly tables added up)", published=int(s["tokyo"][LAST]), published_moe90=None, reproduced=int(H.annual(mt, "starts")[LAST]), kind="count", tolerance=0) A.reproduce("London net additional dwellings, 2024-25 (Table 118, rounded)", published=32680, published_moe90=None, reproduced=int(round(s["london"][2024], -1)), kind="count", tolerance=0) A.reproduce("Toronto area homes started, 2025 (CMHC)", published=26087, published_moe90=None, reproduced=int(s["toronto"][LAST]), kind="count", tolerance=0) A.results["reproduction_source"] = ("MLIT's annual table; MHCLG Table 118's rounded release; CMHC's 2025 total as " "Statistics Canada's table gives it") A.results["reproduction_note"] = "Administrative counts: the files must give the published totals exactly." # ---- printed figures D = A.results["decade"] for k in ("nyc_permits", "nyc_completions", "tokyo", "toronto", "paris", "london", "london_bc_starts"): A.printed(f"decade.{k}.per_1000", f"{D[k]['per_1000']:.1f}") A.printed("decade_times.tokyo_vs_nyc", f"{A.results['decade_times']['tokyo_vs_nyc']:.1f}") A.printed("decade.paris_authorised.per_1000", f"{D['paris_authorised']['per_1000']:.1f}") A.printed("london_ex_adjustment", f"{A.results['london_ex_adjustment']:.1f}") for k in ("tokyo_wards", "toronto_city", "paris_city", "paris_inner"): A.printed(f"core_decade.{k}", f"{A.results['core_decade'][k]:.1f}") A.printed("early_decade.nyc_permits", f"{A.results['early_decade']['nyc_permits']:.1f}") A.printed("nyc_completions_2025.permitted_2022_pct", f"{A.results['nyc_completions_2025']['permitted_2022_pct']:.0f}%") Lg = A.results["long"] for k in ("nyc_peak", "tokyo_peak_year_rate", "nyc_2005_2008_min", "nyc_2005_2008_max", "paris_2005_2008_min", "paris_2005_2008_max", "nyc_2009", "tokyo_1990", "tokyo_2025", "nyc_2025", "nyc_1990s_max", "tokyo_vs_nyc_2025"): A.printed(f"long.{k}", f"{Lg[k]:.1f}") A.printed("long.nyc_peak_units", f"{Lg['nyc_peak_units']:,}") A.printed("long.nyc_2025_units", f"{Lg['nyc_2025_units']:,}") A.printed("long.tokyo_2025_units", f"{Lg['tokyo_2025_units']:,}") Y = A.results["y2025"] A.printed("y2025.nyc_completions.per_1000", f"{Y['nyc_completions']['per_1000']:.1f}") A.printed("y2025.toronto.per_1000", f"{Y['toronto']['per_1000']:.1f}") A.printed("y2025.paris.per_1000", f"{Y['paris']['per_1000']:.1f}") A.printed("nyc_completions_2025.net", f"{A.results['nyc_completions_2025']['net']:,}") N = A.results["now"] for k in ("nyc", "tokyo", "toronto", "paris"): A.printed(f"now.{k}.change_pct", f"{abs(N[k]['change_pct']):.0f}%") for k in ("nyc", "tokyo", "toronto", "paris"): A.printed(f"now.{k}.now", f"{N[k]['now']:,}") A.printed(f"now.{k}.before", f"{N[k]['before']:,}") A.printed("now_times.nyc_lead_over_toronto_pct", f"{A.results['now_times']['nyc_lead_over_toronto_pct']:.0f}%") A.printed("tokyo_march_2025.starts", f"{A.results['tokyo_march_2025']['starts']:,}") A.printed("tokyo_march_2025.median_other_months", f"{A.results['tokyo_march_2025']['median_other_months']:,}") A.printed("nyc_completions_2025.known_permit_year_pct", f"{A.results['nyc_completions_2025']['known_permit_year_pct']:.1f}%") # the page: 2025's permits were fewer than in any year from 2022 to 2024 assert all(s["nyc_permits"][LAST] < s["nyc_permits"][t] for t in (2022, 2023, 2024)) A.printed("now.nyc.per_1000", f"{N['nyc']['per_1000']:.1f}") A.printed("now.toronto.per_1000", f"{N['toronto']['per_1000']:.1f}") A.printed("now.paris.per_1000", f"{N['paris']['per_1000']:.1f}") A.printed("now.tokyo.per_1000", f"{N['tokyo']['per_1000']:.1f}") A.printed("now_times.tokyo_vs_nyc", f"{A.results['now_times']['tokyo_vs_nyc']:.1f}") A.printed("london_undercount.new_build", f"{A.results['london_undercount']['new_build']:,}") A.printed("london_undercount.bc_completions", f"{A.results['london_undercount']['bc_completions']:,}") A.printed("history.coy_homes", f"{A.results['history']['coy_homes']:,}") A.printed("history.coy_years", f"{A.results['history']['coy_years']}") A.results["per_base"] = 1000 A.printed("per_base", "1,000") # ---- charts def cell(k): v = per[k] if k == "london": return {str(t + 1): r(v[t], 2) for t in v.index if FIRST <= t + 1 <= LAST} # a financial year at the year it ends return {str(t): r(v[t], 2) for t in v.index if FIRST <= t <= LAST} A.chart("long_view", lines( title="Tokyo started more homes per resident than New York permitted in every year since 1990", subtitle="New homes a year per 1,000 residents, each city's own measure", x=[[str(t), str(t)] for t in years], series=[[k, label, label.split(",")[0]] for k, label in MAIN], cells={"": {k: cell(k) for k, _ in MAIN}}, format="dec1", axis="New homes per 1,000 residents", x_ticks=[str(t) for t in years if t % 5 == 0 or t == LAST], note="New York: homes in buildings permitted (Census Bureau). Tokyo: homes started, from the notices owners file " "(MLIT). Toronto: homes started in the metropolitan area, from 2001 (CMHC). Paris region: homes started in " "Île-de-France, from 2002 (SDES). London: net additional homes (completions less demolitions), each financial " "year (April to March) drawn at the year it ends, from 2000-01 (MHCLG).", provenance="Census Bureau; MLIT; CMHC via Statistics Canada; SDES; MHCLG; each country's population estimates")) dec_rows = [("nyc_permits", "New York, permitted", "NYC permitted"), ("nyc_completions", "New York, completed (net)", "NYC completed"), ("london", "London, net added", "London net"), ("paris", "Paris region, started", "Paris region"), ("toronto", "Toronto area, started", "Toronto area"), ("tokyo", "Tokyo, started", "Tokyo")] early = {k: per[k].loc[2005:2014] if k == "london" else per[k].loc[2006:2015] for k, _, _ in dec_rows} A.chart("decade", strip( title="New York added the fewest homes per resident in both decades", subtitle="New homes a year per 1,000 residents, averaged over each decade", x=[["2006-2015", "2006 to 2015"], ["2016-2025", "2016 to 2025"]], series=[[k, label, short] for k, label, short in dec_rows], cells={"": {k: {**({"2006-2015": r(early[k].mean(), 2)} if early[k].notna().sum() == 10 else {}), "2016-2025": D[k]["per_1000"]} for k, _, _ in dec_rows}}, format="dec1", axis="New homes per 1,000 residents", note="Each city's own measure (the measures table sets them side by side). New York's completions start in 2010, so their earlier " "decade is left out; London's decades are financial years ending in each span.", provenance="Census Bureau; NYC Department of City Planning; MHCLG; SDES; CMHC; MLIT")) A.chart("this_year", strip( title="New York's permits nearly doubled in the first nine months of 2026", subtitle="New homes in January to September of 2025 and 2026", x=[["2025", "2025"], ["2026", "2026"]], series=[["nyc", "New York, permitted", "New York"], ["tokyo", "Tokyo, started", "Tokyo"], ["toronto", "Toronto area, started", "Toronto"], ["paris", "Paris region, started", "Paris region"]], cells={"": {k: {"2025": N[k]["before"], "2026": N[k]["now"]} for k in ("nyc", "tokyo", "toronto", "paris")}}, format="num", axis="New homes", note="Counts, not rates: the cities differ in size. New York's 2026 permits are preliminary. Tokyo's 2025 includes an " "unusually large March. The Paris region's 2026 comes from the ministry's new system (Sitadel3) from March, " "which slightly revised the published figures.", provenance="Census Bureau; MLIT; CMHC via Statistics Canada; SDES")) rows_out = [] for k, v in s.items(): for t, n in v.items(): pr = per[k].get(t) rows_out.append({"measure": k, "year": int(t), "homes": float(n), "per_1000": None if pr is None else r(pr, 3)}) A.write(data=pd.DataFrame(rows_out)) print(f"{SLUG}: wrote {A.out}") return 0 if __name__ == "__main__": raise SystemExit(main())