compute
ChatGPTStatistics on up to 8 series computed by the server, returning a small JSON instead of rows: resample (freq D, W or M; agg mean or last), transform (level, diff, pct, log, yoy_pct, yoy_diff, auto = the dictionary's change unit), and stats — summary (in every answer), corr (correlation matrix; max_lag adds leads and lags of the others against the first id), regress (OLS of the first id on the others, x lagged by lag periods, Newey-West HAC t), events (average change of each series from the period before to each period of window around the BOK MPC (bok_mpc) or FOMC (fomc) decisions — decision filters hike, cut, hold or change — or around event_dates), series (the resampled, transformed values, ≤ 600 rows). Monthly ids (get_series) and daily ids (get_daily) mix at freq M.
describe_variable
ChatGPTThe full dictionary entry of one variable (definition, unit of its change, source, licence, coverage) with its last 13 monthly values and its daily twin if one exists.
find_apartments
ChatGPTFind Seoul apartment complexes in the 국토교통부 apartment sales records by name (Korean; spaces and the word 아파트 are ignored, part of the name is enough, and the 동 may lead it: 잠실주공5단지, 압구정현대), optionally within a district (gu: 강남구 or 강남) or a 법정동 (dong: 대치동). Returns ids with the district, the 동, the year built, the number of sales since 2006 (direct = how many were 직거래, marked from 2021-11) and the exclusive-area types sold. With only gu or dong it lists that area's complexes, the most traded first. Use an id with get_apartment_trades.
get_apartment_trades
ChatGPTSales of one Seoul apartment complex (an id from find_apartments) in the 국토교통부 records (cancelled deals left out), price in 만원: per month (or quarter, year) and exclusive-area type, the number of sales, the median price, the median price per ㎡, the low and the high; then the latest individual sales (date, exclusive area, floor, price, deal: brokered 중개거래 or direct 직거래 — marked from 2021-11, blank before). Without deal the table counts brokered sales (unmarked ones included, as in Macro Lab's monthly kr_apt_* series) and the list shows every sale; deal brokered, direct or all filters both — 직거래 are often below the market (between relatives), so read them apart. Default: the last 120 months and the latest 20 sales; narrow with from/to (YYYY-MM), area (one type: 84 = 84.00–84.99㎡) and period. At most 600 rows and 200 sales per call. The latest months are still being reported (a sale is reported within 30 days of the contract).
get_daily
ChatGPTDaily or weekly values of up to 8 public daily series (ids from list_datasets or describe_variable daily_twin: DGS10, DFF, KR_CD91, KR_KOSPI …) as CSV. Default: the last 365 rows; narrow with from/to (YYYY-MM-DD) or last (rows). A row is a date any of the series has; align common keeps only the dates all of them have (DFF and KR_BASE run seven days a week, markets on trading days). At most 600 rows per call.
get_decisions
ChatGPTEvery Bank of Korea monetary policy board (금통위) and FOMC rate decision since 2001 — hold included — as CSV or JSON: date, decision (hike, cut, hold; blank for a meeting still to come), change in bp, the rate after it (BOK base rate; the FOMC target or the upper bound of its range, with lower), unscheduled meetings, the effective date when it differs, status (decided, scheduled, pending). Default: the last 10 years of both banks; narrow with bank, from/to, decision.
get_series
ChatGPTMonthly values of up to 12 variables as CSV (date + ids) or JSON. Default: the last 120 months; narrow with from/to (YYYY-MM) or last (months). At most 600 rows and 8000 cells per call — ask again for another window rather than everything at once. Months after a series' last panel value are added from its daily twin when the twin already has them (computed the way the panel takes them, marked).
latest
ChatGPTThe latest monthly value of each variable and, where a public daily twin exists, its latest daily value. Up to 40 ids; none = the key indicators.
list_datasets
ChatGPTWhat the data holds: its version, the period of the monthly panel and its series by group, the daily series ids, the BOK and FOMC decision tables, the licence and how to read the data. Call this first.
search_variables
ChatGPTSearch the data dictionary by words (Korean or English), id, group or country. Returns ids with labels, units, coverage and nature. Use the ids with get_series.