business_cycle: read four data-lectures snapshots and re-enable the lecture

[mmcky]

Re-enables business_cycle on this site. The lecture was left out because its data reads were live API calls, wbgapi against the World Bank and pandas_datareader against FRED. It now reads four cached snapshots that QuantEcon/data-lectures serves and refreshes by pull request, so it can run in the reader's browser. This is block B of the wasm-dynamic-snapshots project, tracked in QuantEcon/data-lectures#118.

Closes #70.

What changes

• lectures/business_cycle.md is a port of lecture-python-intro main's copy, not an edit of this repo's 2025-10-23 copy. The port is from intro commit 901462c; the file last changed there in Update business cycle data to latest available (#364) lecture-python-intro#770. So it carries intro's later changes and its "to the present" prose, which suits snapshots that refresh. That is option (a) of decision 1 on lecture-wasm dynamic snapshots: business_cycle reads the four data-lectures snapshots (P4) data-lectures#118, taken on 2026-09-30. This repo's conventions from ENH: Migrate to use mystmd and quantecon-theme.  #31 are kept: :label: options with :::{figure} blocks, short-form tags and the front matter. Beyond those conventions, only data access changes:
  • the pip install cell and both API imports go, and pyodide_http.patch_all() runs in the imports cell, as in mle and simple_linear_regression;
  • each wb.data.DataFrame(…, labels=True) becomes pd.read_csv(…, index_col='economy'), followed by .loc[[…]] on the economies the API call named;
  • each web.DataReader(id, 'fred', start, end) becomes fred_series(id, start, end), a helper described below;
  • the two wbgapi teaching calls, wb.series.info and wb.series.metadata.get, become non-executing python blocks, with a note linking the intro edition that runs them live;
  • the two year-on-year changes pass fill_method=None to pct_change, described below.
• lectures/myst.yml uncomments business_cycle, between long_run_growth and inflation_history.
• README.md: business_cycle and short_path leave the unsupported list, so the list matches the six entries still commented out in myst.yml (short_path has been enabled in myst.yml since ENH: Migrate to use mystmd and quantecon-theme.  #31). The rule "do not edit lectures here; sync from intro's wasm branch" gives way to "edit lectures/ directly and check the change on the PR's preview". That branch has not changed since 2025-04-24 (c870cca) and is 131 commits behind intro main, and PRs here already edit lectures/ directly.

Enabling the page also mends a link. On main, long_run_growth's [business cycle](business_cycle) builds as a download of the raw .md file; on this branch it links to the page. Nothing is deleted, because this repo holds no copy of these data.

The four snapshots

All four are read at main on raw.githubusercontent.com, the form that works in the browser (data-lectures repoint rule 5).

URL	Source	Read at
https://raw.githubusercontent.com/QuantEcon/data-lectures/main/lectures/gdp_growth_annual.csv	World Bank NY.GDP.MKTP.KD.ZG, nine economies	lines 78, 102 and 503
https://raw.githubusercontent.com/QuantEcon/data-lectures/main/lectures/unemployment_rate_annual.csv	World Bank SL.UEM.TOTL.NE.ZS, four economies	line 568
https://raw.githubusercontent.com/QuantEcon/data-lectures/main/lectures/private_credit_to_gdp.csv	World Bank FS.AST.PRVT.GD.ZS, United Kingdom	line 746
https://raw.githubusercontent.com/QuantEcon/data-lectures/main/lectures/us_business_cycle_monthly.csv	FRED M0892AUSM156SNBR, UNRATE, USREC, UMCSENT, CPILFESL and INDPRO on one monthly index	line 319, then fred_series at eight call sites

For the lecture's authors to accept

The fred_series helper

This is the one piece of new logic that intro does not have (lines 322 to 328):

def fred_series(series_id, start, end):
    """
    One series from the snapshot, between start and end, without the
    empty months before its first observation or after its last.
    """
    data = fred.loc[start:end, [series_id]]
    return data.loc[data.first_valid_index():data.last_valid_index()]

The composite pads all six series to one index starting in 1919, but web.DataReader returned only a series' own observations in the window. The helper returns the same thing. I compared it live with web.DataReader (pandas-datareader 0.11.1) at all eight call sites on 2026-09-30. The frames have the same columns, index name and dtype, value dtypes and null placement, including UNRATE's 2025-10 hole, which FRED also serves. Values differ only where FRED has revised INDPRO since the snapshot was taken, by at most 0.13 in the YoY change, and by FRED's newer month.

The alternative is to repeat the slice and the trim at each of the eight call sites. The helper has one limit, which no call here reaches: for a window with no observations it returns every row of the window, all NaN, rather than an empty frame.

fill_method=None in the two pct_change calls (lines 645 and 704)

Pyodide 0.27.0, which this site loads, ships pandas 2.2.3. There, pct_change(12) pads FRED's 2025-10 hole in CPILFESL forward. That plots a point intro's figure does not have and prints a FutureWarning under the consumption figure. Intro builds on pandas 3 (anaconda 2026.07), where fill_method=None is already the behaviour, so intro does neither. Passing it explicitly makes the browser's figure match intro's, and pandas 3 accepts it unchanged. INDPRO has no hole, so the production figure does not change; the keyword is there for symmetry. Drop both if you would rather keep intro's code verbatim, at the cost of the padded point and the warning.

For an author's eye: five sentences

Two sentences are new, in Data acquisition:

• line 59: "This version of the lecture runs in your browser and reads cached snapshots of the World Bank and FRED data from QuantEcon/data-lectures." (linked to that repo's catalog)
• line 61: "The two calls to the World Bank's data API wbgapi below are shown without running; the edition of this lecture on intro.quantecon.org runs them live." (linked to intro's Data acquisition section)

Three existing sentences are reworded, because the port made them untrue:

• line 47: "We pull all time series through to the most recent available data." now reads "We pull all time series through to the most recent data in the snapshots we read below." A snapshot is refreshed by pull request, so it can trail its source by a month.
• line 74: "Now we use this series ID to obtain the data." now reads "Now we read a snapshot of this series." The code reads a file, not the ID.
• line 85: "We can look at the series' metadata to learn more about the series (click to expand)." now ends at "series.", because the block no longer runs and has no hidden output to expand.

The first non-executing block also gains import wbgapi as wb, because wb is otherwise never bound on the page. The code comment at line 101 now reads "Read the same snapshot again".

What was verified, and how (2026-09-30)

Check	How	Result
Every code cell runs	All 22 {code-cell} blocks of the committed file, in order, in one namespace, under the Pyodide 0.27.0 pins that this build's thebe bundle loads (pandas 2.2.3, numpy 2.0.2, matplotlib 3.8.4; CPython 3.12.13, where Pyodide has 3.12.7). Agg backend, a no-op pyodide_http stub, and imports of wbgapi and pandas_datareader blocked, since the browser has neither	No errors and no warnings. 12 figures, all with data. The only network reads are the four URLs above
The site builds	npx mystmd@1.9.1 build --html in lectures/, as ci.yml does, on this branch rebased on main (ff9196d, theme v3.2.0)	Exit 0. The only new message is the "Extension inferred for table of contents entry" notice that every toc entry gets. The page has 22 executable cells, the two static blocks as plain code, and 12 numbered figures (us_gdp to dcpc)
data-url-guard	Its run: step, taken from .github/workflows/data-url-guard.yml and run on this branch	Passes
URL form	Every URL in a code cell, plus data-lectures' own audit scan (scripts/build_audit.py at main) applied to the file	Six reads of four files, all https://raw.githubusercontent.com/QuantEcon/data-lectures/main/lectures/<file>, with no github.com/…/raw/ form, no media host and no other ref. The audit finds four data-lectures refs and no live-API marker; on main it flags wbgapi and pandas_datareader
The URLs serve	curl with Origin: https://quantecon.github.io, and again with a Netlify preview origin	Each returns HTTP 200 with access-control-allow-origin: *, with bytes matching its manifest's sha256 on data-lectures main at the time
World Bank reads	Live wb.data.DataFrame(…, labels=True) at all four call sites, against the snapshot reads	Same columns, Country labels, nulls and values; only the row order differs
fred_series	The live comparison above	As described above
No conflict with #81	git merge-tree against #81's head	Merges cleanly

Not verified: a run in a browser. The code ran on CPython with pyodide_http stubbed, so the browser's fetch path is untested here.

The gate: Run all on this PR's preview

Someone needs to open this PR's own preview at https://pr-<N>--beautiful-dodol-cb9543.netlify.app/business-cycle/, where N is this PR's number (the "Netlify Preview Ready!" comment gives the URL), and run "Run all". JupyterLite stops at the first failing cell, so the lecture passes when all 12 figures draw. Ignore the red Netlify app check: it fails on every PR here (#62), while the workflow's own preview serves real pages. CI cannot do this check, because ci.yml builds without running any code. Run it after data-lectures' first monthly refresh has merged (its PR should open on Monday 2026-10-05), so the gate tests the bytes this page will read.

Before merging

• The licence record lands first. licence: rest UMCSENT on Michigan's FAQ and record FRED's caching clause data-lectures#152 records the UMCSENT reading, decision 2 on lecture-wasm dynamic snapshots: business_cycle reads the four data-lectures snapshots (P4) data-lectures#118, taken on 2026-09-30: UMCSENT stays permitted on Michigan's FAQ, so the Consumption subsection stays as it is.
• The first monthly refresh lands first, and the gate above runs after it.
• Merge on the same day as block C, Flip wave P4 to repointed — the four business_cycle snapshots data-lectures#153, which records this repo as the four snapshots' consumer and marks them repointed. Between the two merges, data-lectures' strict audit fails, because a landed record has a reader. Every push to data-lectures main and any PR there that touches a manifest go red, and its Pages deploy is skipped. So merge B and C the same day, with no Monday 05:17 UTC audit run between the two merges.

Open questions

• Should update_lectures.py go? This PR keeps it. The README now retires the sync, but the script still downloads intro's wasm branch and copies its lectures/ over this repo's (shutil.copytree(…, dirs_exist_ok=True)). That branch still has a business_cycle.md, so running the script would bring back the old live-API copy, along with other stale files. Delete the script, or keep it as history? Without it, porting from intro means applying its conversions by hand (!pip to %pip, dropping --upgrade).
• Could the two wbgapi calls run here after all? Include business_cycle: what it reads, what data-lectures now serves, and the sequence #70 assumed the API hosts send no CORS headers. Today api.worldbank.org/v2 answers access-control-allow-origin: *, though FRED's fredgraph.csv sends none. Whether wbgapi installs through piplite and runs under pyodide_http is untested, so this PR follows Include business_cycle: what it reads, what data-lectures now serves, and the sequence #70 and shows the calls without running them. Its prose does not claim they cannot run.
• Michigan's citation on the page. data-lectures rests UMCSENT's licence on Michigan's FAQ, which asks users to cite "University of Michigan, Survey Research Center, Surveys of Consumers". The Consumption subsection names "the University of Michigan" and links FRED, as intro does. Adding Michigan's form of the citation there is the authors' call.
• Row order, cosmetic: the two GDP tables list the economies in the lecture's order (USA, ARG, GBR, GRC, JPN), while live wbgapi returns the API's order. Values, labels and figures are unaffected.

Left alone on purpose

This is prose inherited from intro that the snapshots do not fully bear out. Changing it is intro's call, not this port's.

• "from 1970 to the present" for UK credit: the series starts in 1960, at 17.34% of GDP.
• Argentina's growth "did not fall during the two developed economy recessions in the 1970s and 1990s": it is -0.03% in 1975, the last year of the shaded 1973 to 1975 band.
• The comment above end_date, "To freeze the data for a reproducible build…": this site builds without executing code, and a fixed date does not freeze a snapshot that is refreshed in place.

🤖 Generated with Claude Code

[netlify]

❌ Deploy Preview for beautiful-dodol-cb9543 failed.

Name	Link
🔨 Latest commit	adaf9bc
🔍 Latest deploy log	https://app.netlify.com/projects/beautiful-dodol-cb9543/deploys/6abc9634f1e0fe0008690634

[github-actions]

📖 Netlify Preview Ready!

Preview URL: https://pr-85--beautiful-dodol-cb9543.netlify.app (adaf9bc)

✨ Browse the preview at the URL above.