Skip to content
Published studies are shown for demonstration purposes, not because they are good or bad research. Demonstrator · items marked Example are invented · what exists today
Search

How do US bank stocks respond to changes in the Federal Reserve's target rate

How this study was produced: every step of the run with the file it wrote, and everything the run used, pinned to the exact version. The address of this page is derived from its content: if anything below changed, the address would change.

DossierDemonstrationVersion 5 of the studyOpen the study
Dossiersha256:ba888f1a3f12d1b11ce1571973d5e71728c554d57c9639198fd8b11e619c02ee
Dossier as JSON

Paper, data and code of the study

  • Paper
    Not public.

  • Data
    Not public.

    Some of the data come from Yahoo Finance. Its terms allow personal use only, so the data stay with the source.Get the data from the source (its terms), or load them yourself with e2er-data yfinance.

    Data from FRED are open to everyone at the source.

    For the other data of the study, ask the authors.

  • Code
    Not public.

Run settings

e2er
version 0.10.0 · code at 5b8d57ae709d
Template
event-study-finance · single pass
Governance
full
AI backend
claude_code
Models
claude_code/claude-haiku-4-5-20251001
Exported
20261005
Sessions
15 sessions, on 2 e2er commits · listed with the workflow
Outcome
Completed · 2026-09-29 15:59 UTCe2er's internal quality review: 6.1 of 10
e2er code
no local changes at run time

Input data

SHA-256 of the data files, as recorded in the study's provenance.

4 data files with their hash values
  • ·data/data.db 4a037d11941f
  • ·data/data_dictionary.json a8f31a3dfa50
  • ·data/data_sources.json 9234dc17b0e6
  • ·data/data_summary.md 7deb10dbd872

Data used

Yahoo Finance (through yfinance) · 3 loads · read 2026-10-04 · personal use only

e2er's connector for it · about this data source

KBE daily prices · read 2026-10-04 · table kbe_prices · source page Cite: Yahoo Finance, KBE daily prices, retrieved through yfinance on 2026-10-04, https://finance.yahoo.com/quote/KBE.
SPY daily prices · read 2026-10-04 · table spy_prices · source page Cite: Yahoo Finance, SPY daily prices, retrieved through yfinance on 2026-10-04, https://finance.yahoo.com/quote/SPY.
XLF daily prices · read 2026-10-04 · table xlf_prices · source page Cite: Yahoo Finance, XLF daily prices, retrieved through yfinance on 2026-10-04, https://finance.yahoo.com/quote/XLF.

Terms Yahoo's terms of use apply. The yfinance project notes that Yahoo's finance data are intended for personal use only; yfinance is not affiliated with Yahoo. Full terms

The source publishes no citation format; e2er records this one.

FRED, Federal Reserve Bank of St. Louis · 1 series · read 2026-10-04 · FRED's terms of use apply

e2er's connector for it · about this data source

DGS2 · read 2026-10-04 · table dgs2 · source page

Terms FRED's terms of use apply. Each series page states its copyright status: public domain (citation requested), copyrighted with citation required, or copyrighted with the owner's permission needed for any use beyond personal use. Full terms

Cite Board of Governors of the Federal Reserve System (US), Market Yield on U.S. Treasury Securities at 2-Year Constant Maturity, Quoted on an Investment Basis [DGS2], retrieved from FRED, Federal Reserve Bank of St. Louis; https://fred.stlouisfed.org/series/DGS2, October 4, 2026.

The researcher's data folder · 1 file · read 2026-09-28 · terms not known to e2er

e2er's connector for it

fomc_announcement_dates.csv · read 2026-09-28 · table fomc_announcement_dates data/fomc_announcement_dates.csv c1e064d256a4

Terms Supplied by the researcher; e2er does not know the terms of these files.

e2er's internal quality review

Six reviewer specialists each score the draft from one angle, on a scale of 0 to 10. The score in the middle is the weighted average of these scores, with the weights in the table.

data 7.0 identification 4.0 literature 5.7 mechanism 5.9 technical 7.0 writing 8.0 6.1 of 10

6.1 of 10

Each reviewer's score and weight
ReviewerScoreWeight
data7.01.25
identification4.01.5
literature5.71
mechanism5.91
technical7.01.5
writing8.00.75

Pre-registration

The design was written down and frozen before the analysis ran. The hash value is the SHA-256 of the file at that moment.

File
design/preregistration.md
Hash value
5615cb04ba2e
Frozen
2026-09-29T12:49:29Z
Plan files
  • event_design.json da30259ba6c5
  • identification_spec.json 3900f8795b1c

The plan files were frozen with the pre-registration; the hash value after each is the SHA-256 of the file at that moment. The pre-registration check compares the files with these hash values.

Workflow (105 steps)

The run in order, from the run's own record. The run was carried out in 15 sessions on 2 e2er commits. How to read it

A session is one e2er process, from a start or a resume to the next stop, and each session names the e2er version and commit its steps ran on. Each specialist step shows the model, the file it wrote and the file's hash value (SHA-256). Where the run did not record the hash value, the step says so: the last step that wrote a file shows the hash value of the file at export, and earlier steps that wrote the same file are marked as superseded. Checks that stopped a step are listed where they happened: 11 here. Steps taken by the researcher (39 here) are marked with a grey dot. Stops, pauses and failures of the run are marked with a ring, at their time.

50 researcher decisions and stops · 39 actions · 11 stops · 0 approved deviations · 0 failures

All 50, by session
Session 1 of 15 2026-09-28 14:56–23:04 UTC · e2er 0.10.0, commit 04ac10260b73 · 9 steps · idea developer, identification strategist, literature scanner and 1 more · 5 researcher actions · Run stopped for the researcher's review (review design)
  1. Initial
  2. idea developer · claude-haiku-4-5-20251001 · 14:57–14:58 design/paper_plan.md (not recorded during the run; superseded by a later step)
  3. identification strategist · claude-haiku-4-5-20251001 · 14:57–14:59 design/identification_strategy.md (not recorded during the run; superseded by a later step)
  4. literature scanner · claude-haiku-4-5-20251001 · 14:57–15:02 design/literature_review.md 0212e09b04a8 (not recorded during the run; hash of the file at export)
  5. data architect · claude-haiku-4-5-20251001 · 15:02–15:03 data/data_dictionary.json (not recorded during the run; superseded by a later step)
  6. Run stopped for the researcher's review (review design) · 15:03
  7. Researcher gave an instruction for review design · 23:04 Follow the approved design exactly. Events: the 31 FOMC announcement dates in the table fomc_announcement_dates; use no other dates. Day 0 is the announcement date; for the Sunday cut of 15 March 2020, day 0 is the next trading day, 16 March 2020. Asset: KBE; market: SPY (Yahoo Finance); XLF as a robustness check. Data tables: kbe_prices, spy_prices, xlf_prices (Yahoo Finance) and dgs2, dfedtaru (FRED). Market model with estimation window [-250,-12] trading days, leaving out days inside other events' windows. Event windows: [-1,+1] and [0,+5]. H1: the mean CAR differs from zero, tested separately for hikes and cuts (two-sided). H2: the CAR is related to the day-0 change in the 2-year Treasury yield (FRED DGS2). Robustness: without the March 2020 emergency cuts and without the March 2023 hike. Open data only (FRED, Yahoo Finance); no CRSP, Compustat or Y-9C.
  8. Researcher edited event_design.json · 23:04 9abb35d559c5 → a2c01232f499
  9. Researcher sent back idea developer · 23:04 Rewrite paper_plan.md to the approved design in the instructions: one question, H1 and H2 only, KBE against SPY, the 2-year yield (DGS2) as the surprise measure, open data only (no CRSP, no individual bank holding companies).
  10. Researcher sent back data architect · 23:04 Rewrite data_dictionary.json with the tables kbe_prices, spy_prices, xlf_prices (Yahoo Finance), dgs2 and dfedtaru (FRED), and fomc_announcement_dates. No individual banks.
  11. Researcher sent back identification strategist · 23:04 Update identification_strategy.md and identification_spec.json to the estimation window [-250,-12] and the event windows [-1,+1] and [0,+5]. Keep event_design.json exactly as it is now: its 31 events, windows and estimation window were set by the researcher. Do not change it.
Session 2 of 15 2026-09-28 23:04–23:15 UTC · e2er 0.10.0, commit 04ac10260b73 · 4 steps · idea developer, data architect, identification strategist · 1 researcher action · Run stopped for the researcher's review (review design)
  1. Initial
  2. idea developer · claude-haiku-4-5-20251001 · 23:04–23:06 Run again after the researcher sent it back design/paper_plan.md 7ed1c4a9899a (not recorded during the run; hash of the file at export)
  3. data architect · claude-haiku-4-5-20251001 · 23:06–23:07 Run again after the researcher sent it back data/data_dictionary.json a8f31a3dfa50 (not recorded during the run; hash of the file at export)
  4. identification strategist · claude-haiku-4-5-20251001 · 23:07–23:07 Run again after the researcher sent it back design/identification_strategy.md (not recorded during the run; superseded by a later step)
  5. Run stopped for the researcher's review (review design) · 23:07
  6. Researcher approved review design · 23:15
Session 3 of 15 2026-09-28 23:15 to 2026-09-29 07:24 UTC · e2er 0.10.0, commit 04ac10260b73 · 7 steps · data analyst · 2 researcher actions · 3 checks failed
  1. Initial
  2. Check: contracts failed, step stopped
    What the check founddata.db: data_dictionary.json declares tables that were not loaded — missing from data.db: spy_prices, kbe_prices, xlf_prices, dgs2, dfedtaru. Tables in data.db: fomc_announcement_dates (31 rows). Load each series with `e2er-data ... --table <name>`.
  3. data analyst · claude-haiku-4-5-20251001 · 23:15–23:19 data/data_summary.md (stopped by the check above; the step ran again)
  4. data analyst · claude-haiku-4-5-20251001 · 23:19–23:24 data/data_summary.md (not recorded during the run; superseded by a later step)
  5. Check: event window failed, step stopped
    What the check found(c) 1 event date(s) are not trading days in spy_prices.date: fomc-2020-03-15 (2020-03-15); (c) 1 event(s) have windows that run past the data (2015-01-02 to 2025-12-30): fomc-2015-12-16; (d) 1 event(s) are on a different day than in the researcher's table: 2020-03-16 in the table, 2020-03-15 in the design; (d) announcement dates moved to the next trading day: 2020-03-15 → 2020-03-16
  6. Check: event window gate failed, run stopped
    What the check found(c) 1 event date(s) are not trading days in spy_prices.date: fomc-2020-03-15 (2020-03-15); (c) 1 event(s) have windows that run past the data (2015-01-02 to 2025-12-30): fomc-2015-12-16; (d) 1 event(s) are on a different day than in the researcher's table: 2020-03-16 in the table, 2020-03-15 in the design
  7. Researcher edited event_design.json · 07:24 a2c01232f499 → da30259ba6c5
  8. Researcher sent back data analyst · 07:24 Load all price and yield series from 2014-01-02 to 2025-12-31 (the approved proposal), so the first event (2015-12-16) has its full estimation window of 250 trading days. Same tables as in data_dictionary.json; data only, no estimation.
Session 4 of 15 2026-09-29 07:24–08:03 UTC · e2er 0.10.0, commit 04ac10260b73 · 3 steps · data analyst · 1 researcher action · 1 check failed · Run stopped for the researcher's review (event window gate)
  1. Run paused · 07:24 interrupted — the server stopped while this paper was running
  2. Initial
  3. Check: contracts failed, step stopped
    What the check founddata_summary.md: data_summary.md must name each loaded table with its actual row count from data.db; missing or wrong for: spy_prices (actual 3017 rows), kbe_prices (actual 3017 rows), xlf_prices (actual 3017 rows)
  4. data analyst · claude-haiku-4-5-20251001 · 07:24–07:27 Run again after the researcher sent it back data/data_summary.md (stopped by the check above; the step ran again)
  5. Run stopped for the researcher's review (event window gate) · 07:27
  6. Researcher sent back data analyst · 08:03 In data_summary.md, name each loaded table with its actual row count from data.db (spy_prices 3017 rows; kbe_prices 3017 rows; xlf_prices 3017 rows; dgs2 2765 rows; dfedtaru 2765 rows; fomc_announcement_dates 31 rows). Don't reload or change any data; data only, no estimation.
Session 5 of 15 2026-09-29 08:03–08:38 UTC · e2er 0.10.0, commit 04ac10260b73 · 2 steps · data analyst · 1 researcher action · Run stopped for the researcher's review (event window gate)
  1. Initial
  2. data analyst · claude-haiku-4-5-20251001 · 08:03–08:04 Run again after the researcher sent it back data/data_summary.md (not recorded during the run; superseded by a later step)
  3. Run stopped for the researcher's review (event window gate) · 08:04
  4. Researcher approved event window gate · 08:38
Session 6 of 15 2026-09-29 08:38–09:15 UTC · e2er 0.10.0, commit 04ac10260b73 · 4 steps · 1 researcher action · 2 checks failed
  1. Initial
  2. Check: event window passed
  3. Check: preregistration failed, step stopped estimation output before the freeze: fig_car_by_direction.pdf, fig_car_distribution.pdf, process_event_study.py
  4. Check: preregister failed, run stopped
    What the check foundestimation output already exists before the pre-registration was frozen: fig_car_by_direction.pdf, fig_car_distribution.pdf, process_event_study.py. Send back the specialist that produced it; the outputs are then moved to set_aside/ and recorded, and this check runs again.
  5. Researcher sent back data analyst · 09:15 Your last run produced estimation output before the pre-registration was frozen (fig_car_by_direction.pdf, fig_car_distribution.pdf, process_event_study.py); these are set aside. Load, clean and describe the data only: no abnormal returns, no CARs, no estimation scripts, no result figures. Keep data_summary.md with the actual row counts.
Session 7 of 15 2026-09-29 09:15–10:42 UTC · e2er 0.10.0, commit 04ac10260b73 · 3 steps · data analyst · 2 researcher actions · Run stopped for the researcher's review (preregister)
  1. Estimation output set aside · 09:15 estimation output found before the pre-registration was frozen fig_car_by_direction.pdf 74591165bfe0fig_car_distribution.pdf 2ebce20b9facprocess_event_study.py ca339babb339
  2. Initial
  3. data analyst · claude-haiku-4-5-20251001 · 09:15–09:18 Run again after the researcher sent it back data/data_summary.md (not recorded during the run; superseded by a later step)
  4. Run stopped for the researcher's review (preregister) · 09:18
  5. Researcher gave an instruction for preregister · 10:42 Disclosure: abnormal returns (event_study_results.csv) were computed by the data analyst at 01:23 on 29 September 2026, eight minutes after the design was approved and before the pre-registration was frozen. The file was set aside unread; the design was not changed after it appeared. The pre-registration check did not recognise result files at that time; this was fixed before the freeze.
  6. Researcher sent back identification strategist · 10:42 Align identification_strategy.md and identification_spec.json with the approved H2: the CAR is regressed on the day-0 change in the 2-year Treasury yield (DGS2) as the explanatory variable, separately reported for hikes and cuts where the design says so. Remove forward_guidance_indicator, which is not in the data. Keep event_design.json unchanged.
Session 8 of 15 2026-09-29 10:42–12:49 UTC · e2er 0.10.0, commit 5b8d57ae709d · 6 steps · identification strategist · 3 researcher actions · Run stopped for the researcher's review (preregister)
  1. Estimation output set aside · 10:42 estimation output found before the pre-registration was frozen event_study_results.csv 3c8769c9797f
  2. Initial
  3. identification strategist · claude-haiku-4-5-20251001 · 10:42–10:45 Run again after the researcher sent it back design/identification_strategy.md 0d74339ae786 (not recorded during the run; hash of the file at export)
  4. Check: preregistration passed
  5. Check: preregistration passed
  6. Run stopped for the researcher's review (preregister) · 10:45
  7. Researcher edited preregistration.md · 12:49 4ad98d083c74 → 5615cb04ba2e
  8. Researcher gave an instruction for preregister · 12:49 The pre-registration as edited at this step governs: H2's primary test is pooled over all 31 events with HC1 standard errors, with hikes (n=20) and cuts (n=11) as a secondary analysis; H1 uses the cross-sectional standard error. Where identification_spec.json's notes say otherwise (e.g. 'n≈17', 'n≈10', separate regressions only), the pre-registration takes precedence.
  9. Researcher approved preregister · 12:49
Session 9 of 15 2026-09-29 12:49–13:30 UTC · e2er 0.10.0, commit 5b8d57ae709d · 13 steps · econometrics specialist, abstract writer, paper drafter · 3 researcher actions · 2 checks failed · Run stopped for the researcher's review (review draft)
Initial 8 steps · econometrics specialist, abstract writer, paper drafter · 12:49–13:04 · 2 checks failed · Pre-registration frozen
  1. Check: preregistration passed
  2. Pre-registration frozen · 12:49 hash value 5615cb04ba2e
  3. Check: contracts failed, step stopped estimation_results.json: empty JSON ('{}')
  4. econometrics specialist · claude-haiku-4-5-20251001 · 12:49–12:50 design/econometric_spec.md (stopped by the check above; the step ran again)
  5. Check: contracts failed, step stopped
    What the check foundestimation_results.json: no estimated regression: estimation_results.json has no non-empty 'coefficients' block (descriptive-only output). Estimate at least one identified specification and write its coefficients/SEs/p-values per the estimation-results-schema — descriptive summaries do not substitute for the estimation.
  6. econometrics specialist · claude-haiku-4-5-20251001 · 12:50–12:55 design/econometric_spec.md (stopped by the check above; the step ran again)
  7. econometrics specialist · claude-haiku-4-5-20251001 · 12:55–12:59 design/econometric_spec.md (not recorded during the run; superseded by a later step)
  8. abstract writer · claude-haiku-4-5-20251001 · 12:59–13:00 paper/abstract.tex (not recorded during the run; superseded by a later step)
  9. paper drafter · claude-haiku-4-5-20251001 · 12:59–13:04 paper/paper.tex (not recorded during the run; superseded by a later step)
Estimation gate 5 steps · 13:04–13:30 · 3 researcher actions · Run stopped for the researcher's review (review draft)
  1. Check: preregistration passed No change since the pre-registration was frozen (2026-09-29 12:49 UTC).
  2. Check: estimation passed
  3. Run stopped for the researcher's review (review draft) · 13:04
  4. Researcher sent back econometrics specialist · 13:30 Use all 31 events from fomc_announcement_dates (20 hikes, 11 cuts), or name and justify any exclusion in estimation_results.json. Compute every p-value from its t-statistic with n-1 degrees of freedom (two-sided). Estimate H2 exactly as pre-registered: CAR[-1,+1] regressed on the day-0 change in DGS2, pooled over all events with HC1 standard errors as the primary test, and separately for hikes and cuts as a secondary analysis; add these results to estimation_results.json. Do not change the pre-registered design.
  5. Researcher sent back paper drafter · 13:30 Update the draft to the corrected estimation results: 31 events, p-values computed from t with n-1 degrees of freedom, and the H2 results (pooled primary, by direction secondary). Report every number from estimation_results.json; state null results plainly as nulls.
  6. Researcher sent back abstract writer · 13:30 Rewrite the abstract so it reports the main results (H1 by direction and H2) with their numbers from estimation_results.json, not only the plan.
Session 10 of 15 2026-09-29 13:30–15:05 UTC · e2er 0.10.0, commit 5b8d57ae709d · 7 steps · econometrics specialist, paper drafter, abstract writer · 4 researcher actions · Run stopped for the researcher's review (review draft)
  1. Estimation gate
  2. econometrics specialist · claude-haiku-4-5-20251001 · 13:30–13:37 Run again after the researcher sent it back design/econometric_spec.md (not recorded during the run; superseded by a later step)
  3. paper drafter · claude-haiku-4-5-20251001 · 13:37–13:41 Run again after the researcher sent it back paper/paper.tex (not recorded during the run; superseded by a later step)
  4. abstract writer · claude-haiku-4-5-20251001 · 13:41–13:41 Run again after the researcher sent it back paper/abstract.tex (not recorded during the run; superseded by a later step)
  5. Run stopped for the researcher's review (review draft) · 13:41
  6. Researcher sent back data analyst · 15:05 The dgs2 table has 2765 rows and no values: the FRED key had a leading space and every request failed. The key is fixed. Reload DGS2 from FRED so that every trading day carries its value, and confirm the non-null count in data_summary.md. If a download fails, stop and say so; never write a table without values. Change nothing else.
  7. Researcher sent back econometrics specialist · 15:05 Compute every p-value with scipy.stats (t distribution with the df you report) or statsmodels; never hand-written distribution functions. scipy and statsmodels are installed. Give each result entry a hypothesis id (H1, H1a, H2, ...) and its df. Estimate H1 as pre-registered on all 31 events (report any exclusion with its reason). Estimate H2 exactly as pre-registered: CAR_i = alpha + beta * dDGS2_0,i with dDGS2 in basis points (FRED reports percent), HC1 standard errors, pooled over all 31 events as the primary test, hikes (n=20) and cuts (n=11) separately as secondary. Report results as they come out, nulls as nulls.
  8. Researcher sent back paper drafter · 15:05 Rewrite the results from the new estimation_results.json only: H1 and H2 as pre-registered, the numbers exactly as estimated, non-significant results stated as such. Keep the demonstration disclaimer.
  9. Researcher sent back abstract writer · 15:05 State the H1 and H2 results as estimated in the new estimation_results.json, including null results.
Session 11 of 15 2026-09-29 15:05–15:16 UTC · e2er 0.10.0, commit 5b8d57ae709d · 8 steps · data analyst, econometrics specialist, paper drafter and 1 more · 4 researcher actions · Run stopped for the researcher's review (review draft)
  1. Estimation gate
  2. data analyst · claude-haiku-4-5-20251001 · 15:05–15:05 Run again after the researcher sent it back data/data_summary.md (not recorded during the run; superseded by a later step)
  3. econometrics specialist · claude-haiku-4-5-20251001 · 15:05–15:12 Run again after the researcher sent it back design/econometric_spec.md (not recorded during the run; superseded by a later step)
  4. paper drafter · claude-haiku-4-5-20251001 · 15:12–15:14 Run again after the researcher sent it back paper/paper.tex (not recorded during the run; superseded by a later step)
  5. abstract writer · claude-haiku-4-5-20251001 · 15:14–15:15 Run again after the researcher sent it back paper/abstract.tex (not recorded during the run; superseded by a later step)
  6. Run stopped for the researcher's review (review draft) · 15:15
  7. Researcher sent back data analyst · 15:16 The previous rerun did not reload anything: data.db table dgs2 still has 2765 rows and 0 values. FRED works now. Run exactly this, which replaces the table: `e2er-data fred series --series-id DGS2 --start 2015-01-01 --end 2025-12-31 --paper-id 9a623c39-87df-4f46-9d6f-d9dcabc84e7a --table dgs2`. Then query data.db and confirm that dgs2 has non-null values on the trading days, and update data_summary.md with the row count and the non-null count. Do not touch the other tables.
  8. Researcher sent back econometrics specialist · 15:16 Keep H1 as it is now (p-values from scipy are correct). Add H2, which is still missing, exactly as pre-registered, using the reloaded dgs2 table: CAR_i[-1,+1] = alpha + beta * dDGS2_0,i, with dDGS2 = DGS2(day 0) - DGS2(previous trading day) in basis points (FRED gives percent, multiply by 100), OLS with HC1 standard errors via statsmodels. Primary: pooled over all 31 events (hypothesis id H2). Secondary: hikes (n=20) and cuts (n=11) separately. Report beta, HC1 SE, t, p, R2, n and df for each.
  9. Researcher sent back paper drafter · 15:16 Update the results from the new estimation_results.json: H1 as it is, and add H2 as estimated. Numbers exactly as estimated, non-significant results stated as such. Keep the demonstration disclaimer.
  10. Researcher sent back abstract writer · 15:16 State the H1 and H2 results as estimated in the new estimation_results.json, including null results.
Session 12 of 15 2026-09-29 15:16–15:27 UTC · e2er 0.10.0, commit 5b8d57ae709d · 8 steps · data analyst, econometrics specialist, paper drafter and 1 more · 4 researcher actions · Run stopped for the researcher's review (review draft)
  1. Estimation gate
  2. data analyst · claude-haiku-4-5-20251001 · 15:16–15:18 Run again after the researcher sent it back data/data_summary.md 7deb10dbd872 (not recorded during the run; hash of the file at export)
  3. econometrics specialist · claude-haiku-4-5-20251001 · 15:18–15:23 Run again after the researcher sent it back design/econometric_spec.md (not recorded during the run; superseded by a later step)
  4. paper drafter · claude-haiku-4-5-20251001 · 15:23–15:25 Run again after the researcher sent it back paper/paper.tex (not recorded during the run; superseded by a later step)
  5. abstract writer · claude-haiku-4-5-20251001 · 15:25–15:26 Run again after the researcher sent it back paper/abstract.tex (not recorded during the run; superseded by a later step)
  6. Run stopped for the researcher's review (review draft) · 15:26
  7. Researcher gave an instruction for review draft · 15:27 The researcher loaded the 2-year Treasury yield with e2er's FRED tool because the data analyst did not, in three attempts: `e2er-data --paper-id 9a623c39-87df-4f46-9d6f-d9dcabc84e7a fred series --series-id DGS2 --start 2015-01-01 --end 2025-12-31 --table dgs2`. data.db table dgs2 now has 2870 rows, 2750 with a value (FRED reports no value on market holidays), 2015-01-01 to 2025-12-31, in percent. The earlier dgs2 table had 2765 rows and no values: a leading space in the FRED key made every request fail, and the analyst's own script wrote the empty rows. No other table was changed.
  8. Researcher sent back econometrics specialist · 15:27 The dgs2 table now has values (see the researcher instruction). Keep H1 exactly as it is. Add H2 exactly as pre-registered: CAR_i[-1,+1] = alpha + beta * dDGS2_0,i, with dDGS2 = DGS2(day 0) - DGS2(previous trading day with a value), in basis points (the table is in percent, multiply by 100); OLS with HC1 standard errors via statsmodels. Primary: pooled over all 31 events, hypothesis id H2. Secondary: hikes (n=20) and cuts (n=11) separately. Report beta, HC1 SE, t, p, R2, n and df for each.
  9. Researcher sent back paper drafter · 15:27 Update the results from the new estimation_results.json: H1 as it is, add H2 as estimated. Numbers exactly as estimated, non-significant results stated as such. State in the data section that the researcher loaded DGS2 after the data analyst failed to. Keep the demonstration disclaimer.
  10. Researcher sent back abstract writer · 15:27 State the H1 and H2 results as estimated in the new estimation_results.json, including null results.
Session 13 of 15 2026-09-29 15:27–15:35 UTC · e2er 0.10.0, commit 5b8d57ae709d · 6 steps · econometrics specialist, paper drafter, abstract writer · 3 researcher actions · Run stopped for the researcher's review (review draft)
  1. Estimation gate
  2. econometrics specialist · claude-haiku-4-5-20251001 · 15:27–15:30 Run again after the researcher sent it back design/econometric_spec.md (not recorded during the run; superseded by a later step)
  3. paper drafter · claude-haiku-4-5-20251001 · 15:30–15:33 Run again after the researcher sent it back paper/paper.tex (not recorded during the run; superseded by a later step)
  4. abstract writer · claude-haiku-4-5-20251001 · 15:33–15:34 Run again after the researcher sent it back paper/abstract.tex (not recorded during the run; superseded by a later step)
  5. Run stopped for the researcher's review (review draft) · 15:34
  6. Researcher sent back econometrics specialist · 15:35 H1 is correct; keep it unchanged. H2: fit with sm.OLS(y, X).fit(cov_type='HC1', use_t=True) so that p-values come from the t distribution with n-2 residual df (the current fit uses statsmodels' normal default, e.g. cuts p=0.460 where t(9) gives 0.479). Add a df field to every result entry (n-1 for the mean tests, n-2 for the regressions) and a hypothesis field holding the id (H1a, H1b, H2, ...). Change nothing else.
  7. Researcher sent back paper drafter · 15:35 Update the H2 numbers from the new estimation_results.json. Report the pre-registered Table 1 in full: hikes and cuts in both windows, [-1,+1] and [0,+5]. Report the pooled H1 estimates as not pre-registered, including the pooled [0,+5] mean CAR of -1.47% (p=0.022). Delete the sentence saying H2 is to be completed once yield data are obtained; H2 is estimated. Do not claim that announcements do not move bank stocks: state what was estimated, which results are significant and which are not. Keep the demonstration disclaimer.
  8. Researcher sent back abstract writer · 15:35 Use the new estimation_results.json. Report the pre-registered results (H1 by direction, H2 pooled) and mention that the pooled [0,+5] estimate, not pre-registered, is -1.47% with p=0.022. No claim that announcements do not affect bank stocks.
Session 14 of 15 2026-09-29 15:35–15:47 UTC · e2er 0.10.0, commit 5b8d57ae709d · 5 steps · econometrics specialist, paper drafter, abstract writer · 2 researcher actions · Run stopped for the researcher's review (review draft)
  1. Estimation gate
  2. econometrics specialist · claude-haiku-4-5-20251001 · 15:35–15:42 Run again after the researcher sent it back design/econometric_spec.md f90b4f6264c9 (not recorded during the run; hash of the file at export)
  3. paper drafter · claude-haiku-4-5-20251001 · 15:42–15:44 Run again after the researcher sent it back paper/paper.tex 0d2d0976485c (not recorded during the run; hash of the file at export)
  4. abstract writer · claude-haiku-4-5-20251001 · 15:44–15:46 Run again after the researcher sent it back paper/abstract.tex 2bd086599815 (not recorded during the run; hash of the file at export)
  5. Run stopped for the researcher's review (review draft) · 15:46
  6. Researcher gave an instruction for review draft · 15:47 Known issues at approval, left as is because this study is published only to demonstrate e2er: (1) The pre-registered robustness checks (without the March 2020 cuts, without the March 2023 hike, XLF as benchmark) are not reported in the paper. Leftover files tables/robustness_march2020.tex, robustness_march2023.tex and xlf.tex are not part of the paper and contradict the pre-registration (three March 2020 cuts instead of two; sample sizes that do not add up). (2) The abstract calls the [0,+5] analysis not pre-registered; the window was pre-registered by direction, only the pooled estimate was not. (3) The 2-year yield series was loaded by the researcher, not by the data analyst (see the earlier instruction). (4) The H2 estimates changed between reruns once the day-0 change was taken against the previous trading day with a value; the final numbers use that definition. The p-values in estimation_results.json were checked against the t distribution with the stated df before approval.
  7. Researcher approved review draft · 15:47
Session 15 of 15 2026-09-29 15:47 to 2026-10-04 22:19 UTC · e2er 0.10.0, commit 5b8d57ae709d · 20 steps · mechanism reviewer, technical reviewer, literature reviewer and 5 more · 3 researcher actions · 3 checks failed
Estimation gate 2 steps · 15:47
  1. Check: preregistration passed No change since the pre-registration was frozen (2026-09-29 12:49 UTC).
  2. Check: estimation passed
Review 12 steps · mechanism reviewer, technical reviewer, literature reviewer and 3 more · 15:47–15:55 · 2 checks failed
  1. Check: numbers passed
  2. Check: citations passed
  3. Check: contracts failed, step stopped review_mechanism.md: file not written
  4. mechanism reviewer · claude-haiku-4-5-20251001 · 15:47–15:49
  5. technical reviewer · claude-haiku-4-5-20251001 · 15:47–15:49 reviews/review_technical.md 4751e5409383 (not recorded during the run; hash of the file at export)
  6. literature reviewer · claude-haiku-4-5-20251001 · 15:47–15:50 reviews/review_literature.md 71f601a0d272 (not recorded during the run; hash of the file at export)
  7. writing reviewer · claude-haiku-4-5-20251001 · 15:49–15:51 reviews/review_writing.md ef97b561f9ad (not recorded during the run; hash of the file at export)
  8. data reviewer · claude-haiku-4-5-20251001 · 15:49–15:51 reviews/review_data.md f2e62f236cc7 (not recorded during the run; hash of the file at export)
  9. identification reviewer · claude-haiku-4-5-20251001 · 15:50–15:53 reviews/review_identification.md c09df1828508 (not recorded during the run; hash of the file at export)
  10. Check: contracts failed, step stopped review_mechanism.md: file not written
  11. mechanism reviewer · claude-haiku-4-5-20251001 · 15:51–15:53
  12. mechanism reviewer · claude-haiku-4-5-20251001 · 15:53–15:55 reviews/review_mechanism.md 886ecc3ba720 (not recorded during the run; hash of the file at export)
Revision 2 steps · patch revisor · 15:55–15:57 · 1 check failed
  1. Check: contracts failed, step stopped paper_draft.tex.edits.json: empty JSON ('[]')
  2. patch revisor · claude-haiku-4-5-20251001 · 15:55–15:57 misc/paper_draft.tex.edits.json (stopped by the check above)
Replication 4 steps · replication packager · 09-29 15:57 to 10-04 22:19 · 3 researcher actions
  1. replication packager · claude-haiku-4-5-20251001 · 15:57–15:59 replication/estimation.py 34deb0cdfb48 (not recorded during the run; hash of the file at export)
  2. Researcher edited data_dictionary.json · 22:19 4937f67ab388 → a8f31a3dfa50 dfedtaru: corrected the origin; it was built from the FOMC dates file on KBE trading days, in basis points, not loaded from FRED
  3. Researcher edited data_summary.md · 22:19 3b0e951ead20 → 7deb10dbd872 dfedtaru: corrected the origin; it was built from the FOMC dates file on KBE trading days, in basis points, not loaded from FRED
  4. Researcher gave an instruction for after approval · 22:19 Sources recorded on 4 October 2026 by reloading with e2er 0.13.5. DGS2 values are identical. Yahoo prices agree to within 1e-6 relative, the rounding by which two Yahoo requests differ; volumes and dividends are identical. dfedtaru was built from the FOMC dates file, not loaded from FRED; its values equal FRED DFEDTARU.

The run ended: completed · 2026-09-29 15:59 UTC

What it was built with (19)

Each part with its authors and the version used. e2er parts are pinned by git's content hash of their file at the commit above; follow a link to read that exact version. A part used on more than one e2er commit lists its version at each commit.

Templates (1)

· by e2er authors · pinned at 2 e2er commits
TemplateByPinned version
event-study-financee2er authors unresolved: pipelines/event-study-finance.toml at 04ac10260b73 is not in the E2ER checkout this dossier was built from · e2er 04ac10260b73unresolved: pipelines/event-study-finance.toml at 5b8d57ae709d is not in the E2ER checkout this dossier was built from · e2er 5b8d57ae709d

Specialists (16)

· by e2er authors · pinned at 2 e2er commits
SpecialistByPinned version
abstract_writere2er authors unresolved: src/core/specialists/registry.py at 5b8d57ae709d is not in the E2ER checkout this dossier was built from
data_analyste2er authors unresolved: src/core/specialists/registry.py at 04ac10260b73 is not in the E2ER checkout this dossier was built from · e2er 04ac10260b73unresolved: src/core/specialists/registry.py at 5b8d57ae709d is not in the E2ER checkout this dossier was built from · e2er 5b8d57ae709d
data_architecte2er authors unresolved: src/core/specialists/registry.py at 04ac10260b73 is not in the E2ER checkout this dossier was built from
data_reviewere2er authors unresolved: src/core/specialists/registry.py at 5b8d57ae709d is not in the E2ER checkout this dossier was built from
econometrics_specialiste2er authors unresolved: src/core/specialists/registry.py at 5b8d57ae709d is not in the E2ER checkout this dossier was built from
idea_developere2er authors unresolved: src/core/specialists/registry.py at 04ac10260b73 is not in the E2ER checkout this dossier was built from
identification_reviewere2er authors unresolved: src/core/specialists/registry.py at 5b8d57ae709d is not in the E2ER checkout this dossier was built from
identification_strategiste2er authors unresolved: src/core/specialists/registry.py at 04ac10260b73 is not in the E2ER checkout this dossier was built from · e2er 04ac10260b73unresolved: src/core/specialists/registry.py at 5b8d57ae709d is not in the E2ER checkout this dossier was built from · e2er 5b8d57ae709d
literature_reviewere2er authors unresolved: src/core/specialists/registry.py at 5b8d57ae709d is not in the E2ER checkout this dossier was built from
literature_scannere2er authors unresolved: src/core/specialists/registry.py at 04ac10260b73 is not in the E2ER checkout this dossier was built from
mechanism_reviewere2er authors unresolved: src/core/specialists/registry.py at 5b8d57ae709d is not in the E2ER checkout this dossier was built from
paper_draftere2er authors unresolved: src/core/specialists/registry.py at 5b8d57ae709d is not in the E2ER checkout this dossier was built from
patch_revisore2er authors unresolved: src/core/specialists/registry.py at 5b8d57ae709d is not in the E2ER checkout this dossier was built from
replication_packagere2er authors unresolved: src/core/specialists/registry.py at 5b8d57ae709d is not in the E2ER checkout this dossier was built from
technical_reviewere2er authors unresolved: src/core/specialists/registry.py at 5b8d57ae709d is not in the E2ER checkout this dossier was built from
writing_reviewere2er authors unresolved: src/core/specialists/registry.py at 5b8d57ae709d is not in the E2ER checkout this dossier was built from

Skills (1)

· by e2er authors · pinned at 2 e2er commits
SkillByPinned version
event-studye2er authors unresolved: skills/files/econometrics/event-study.md at 04ac10260b73 is not in the E2ER checkout this dossier was built from · e2er 04ac10260b73unresolved: skills/files/econometrics/event-study.md at 5b8d57ae709d is not in the E2ER checkout this dossier was built from · e2er 5b8d57ae709d

Connectors (1)

· by e2er authors · pinned at 2 e2er commits
ConnectorByPinned version
claude-codee2er authors unresolved: src/modules/llm/claude_code.py at 04ac10260b73 is not in the E2ER checkout this dossier was built from · e2er 04ac10260b73unresolved: src/modules/llm/claude_code.py at 5b8d57ae709d is not in the E2ER checkout this dossier was built from · e2er 5b8d57ae709d

Cite e2er with this dossier

A paper produced with e2er gives one reference for e2er: this dossier. It credits everyone listed above.

@misc{e2er-ba888f1a3f12d1b1,
  title        = {e2er dossier for "How do US bank stocks respond to changes in the Federal Reserve's target rate"},
  author       = {{e2er}},
  year         = {2026},
  howpublished = {\url{https://e2er.org/d/ba888f1a3f12d1b1}},
  note         = {Workflow, templates, specialists, skills, connectors and models used, pinned by content hash. Dossier sha256:ba888f1a3f12d1b11ce1571973d5e71728c554d57c9639198fd8b11e619c02ee}
}