Upload README.md with huggingface_hub

README.md

@@ -1,176 +1,231 @@
 ---
-dataset_info:
-  features:
-  - name: qid
-    dtype: string
-  - name: forecastType
-    dtype: string
-  - name: subtype
-    dtype: string
-  - name: indicator
-    dtype: string
-  - name: transform
-    dtype: string
-  - name: target_period
-    dtype: string
-  - name: info_cutoff
-    dtype: string
-  - name: question
-    dtype: string
-  - name: options
-    list: string
-  - name: answer_letter
-    dtype: string
-  - name: answer_raw
-    dtype: string
-  - name: unit
-    dtype: string
-  - name: year
-    dtype: int32
-  splits:
-  - name: '1999'
-    num_bytes: 101080
-    num_examples: 338
-  - name: '2000'
-    num_bytes: 121750
-    num_examples: 400
-  - name: '2001'
-    num_bytes: 126277
-    num_examples: 400
-  - name: '2002'
-    num_bytes: 121095
-    num_examples: 400
-  - name: '2003'
-    num_bytes: 122754
-    num_examples: 400
-  - name: '2004'
-    num_bytes: 121268
-    num_examples: 400
-  - name: '2005'
-    num_bytes: 120224
-    num_examples: 400
-  - name: '2006'
-    num_bytes: 121284
-    num_examples: 400
-  - name: '2007'
-    num_bytes: 124130
-    num_examples: 400
-  - name: '2008'
-    num_bytes: 127446
-    num_examples: 400
-  - name: '2009'
-    num_bytes: 121940
-    num_examples: 400
-  - name: '2010'
-    num_bytes: 122145
-    num_examples: 400
-  - name: '2011'
-    num_bytes: 122298
-    num_examples: 400
-  - name: '2012'
-    num_bytes: 121627
-    num_examples: 400
-  - name: '2013'
-    num_bytes: 122768
-    num_examples: 400
-  - name: '2014'
-    num_bytes: 121219
-    num_examples: 400
-  - name: '2015'
-    num_bytes: 123306
-    num_examples: 400
-  - name: '2016'
-    num_bytes: 122063
-    num_examples: 400
-  - name: '2017'
-    num_bytes: 120822
-    num_examples: 400
-  - name: '2018'
-    num_bytes: 122687
-    num_examples: 400
-  - name: '2019'
-    num_bytes: 120861
-    num_examples: 400
-  - name: '2020'
-    num_bytes: 126807
-    num_examples: 400
-  - name: '2021'
-    num_bytes: 121156
-    num_examples: 400
-  - name: '2022'
-    num_bytes: 125780
-    num_examples: 400
-  - name: '2023'
-    num_bytes: 123066
-    num_examples: 400
-  - name: '2024'
-    num_bytes: 124307
-    num_examples: 400
-  - name: '2025'
-    num_bytes: 120009
-    num_examples: 400
-  - name: '2026'
-    num_bytes: 39771
-    num_examples: 130
-  download_size: 1287875
-  dataset_size: 3329940
+language: en
+license: cc-by-4.0
+task_categories:
+  - multiple-choice
+  - question-answering
+size_categories:
+  - 10K<n<100K
+pretty_name: FinDeepForecast-Historical-US
+tags:
+  - finance
+  - economics
+  - macro
+  - forecasting
+  - federal-reserve
+  - time-series
+  - benchmark
 configs:
-- config_name: default
-  data_files:
-  - split: '1999'
-    path: data/1999-*
-  - split: '2000'
-    path: data/2000-*
-  - split: '2001'
-    path: data/2001-*
-  - split: '2002'
-    path: data/2002-*
-  - split: '2003'
-    path: data/2003-*
-  - split: '2004'
-    path: data/2004-*
-  - split: '2005'
-    path: data/2005-*
-  - split: '2006'
-    path: data/2006-*
-  - split: '2007'
-    path: data/2007-*
-  - split: '2008'
-    path: data/2008-*
-  - split: '2009'
-    path: data/2009-*
-  - split: '2010'
-    path: data/2010-*
-  - split: '2011'
-    path: data/2011-*
-  - split: '2012'
-    path: data/2012-*
-  - split: '2013'
-    path: data/2013-*
-  - split: '2014'
-    path: data/2014-*
-  - split: '2015'
-    path: data/2015-*
-  - split: '2016'
-    path: data/2016-*
-  - split: '2017'
-    path: data/2017-*
-  - split: '2018'
-    path: data/2018-*
-  - split: '2019'
-    path: data/2019-*
-  - split: '2020'
-    path: data/2020-*
-  - split: '2021'
-    path: data/2021-*
-  - split: '2022'
-    path: data/2022-*
-  - split: '2023'
-    path: data/2023-*
-  - split: '2024'
-    path: data/2024-*
-  - split: '2025'
-    path: data/2025-*
-  - split: '2026'
-    path: data/2026-*
+  - config_name: default
+    data_files:
+      - split: "1999"
+        path: "data/1999-*.parquet"
+      - split: "2000"
+        path: "data/2000-*.parquet"
+      - split: "2001"
+        path: "data/2001-*.parquet"
+      - split: "2002"
+        path: "data/2002-*.parquet"
+      - split: "2003"
+        path: "data/2003-*.parquet"
+      - split: "2004"
+        path: "data/2004-*.parquet"
+      - split: "2005"
+        path: "data/2005-*.parquet"
+      - split: "2006"
+        path: "data/2006-*.parquet"
+      - split: "2007"
+        path: "data/2007-*.parquet"
+      - split: "2008"
+        path: "data/2008-*.parquet"
+      - split: "2009"
+        path: "data/2009-*.parquet"
+      - split: "2010"
+        path: "data/2010-*.parquet"
+      - split: "2011"
+        path: "data/2011-*.parquet"
+      - split: "2012"
+        path: "data/2012-*.parquet"
+      - split: "2013"
+        path: "data/2013-*.parquet"
+      - split: "2014"
+        path: "data/2014-*.parquet"
+      - split: "2015"
+        path: "data/2015-*.parquet"
+      - split: "2016"
+        path: "data/2016-*.parquet"
+      - split: "2017"
+        path: "data/2017-*.parquet"
+      - split: "2018"
+        path: "data/2018-*.parquet"
+      - split: "2019"
+        path: "data/2019-*.parquet"
+      - split: "2020"
+        path: "data/2020-*.parquet"
+      - split: "2021"
+        path: "data/2021-*.parquet"
+      - split: "2022"
+        path: "data/2022-*.parquet"
+      - split: "2023"
+        path: "data/2023-*.parquet"
+      - split: "2024"
+        path: "data/2024-*.parquet"
+      - split: "2025"
+        path: "data/2025-*.parquet"
+      - split: "2026"
+        path: "data/2026-*.parquet"
 ---
+
+# FinDeepForecast-Historical-US
+
+A historical multiple-choice extension of the [FinDeepForecast](https://openfinarena.com/fin-deep-forecast/) benchmark from OpenFinArena.
+While the original FinDeepForecast is a **live** benchmark whose ground truth can
+only be revealed by the future, this dataset uses **realized** historical values
+from 1999 to 2026 as ground truth, enabling reproducible offline evaluation.
+
+## Highlights
+
+- **10,868 multiple-choice questions** across **28 years** (1999–2026)
+- **49 US macro / market indicators** sourced from [FRED](https://fred.stlouisfed.org/)
+  (St. Louis Fed) — CPI, GDP, Fed Funds, Treasury yields, NASDAQ, VIX, oil, FX, etc.
+- **Two question types** (aligned with the original paper):
+  - **Recurrent** — forecast periodic data releases (CPI for next month, GDP for next quarter, etc.)
+  - **Situational** — non-periodic queries: yearly stats, trends, and event-driven
+    questions for ~35 hand-curated macro events (FOMC decisions, NBER recessions,
+    Lehman, COVID, SVB, etc.)
+- **Year-as-split** layout: each year is a separate split for curriculum / leak-free evaluation
+
+## Quick Start
+
+```python
+from datasets import load_dataset
+
+# Load a single year
+ds = load_dataset("TheFinAI/pre_test", split="2008")
+print(ds[0])
+
+# Load all years
+ds = load_dataset("TheFinAI/pre_test")
+print(ds)  # DatasetDict with keys "1999"..."2026"
+```
+
+## Distribution
+
+| forecastType | subtype                  | Count |  %   |
+|--------------|--------------------------|------:|-----:|
+| Recurrent    | recurrent                | 6,425 | 59.1 |
+| Situational  | trend                    | 1,202 | 11.1 |
+| Situational  | window_max               | 968   |  8.9 |
+| Situational  | window_min               | 966   |  8.9 |
+| Situational  | yearly_return            | 537   |  4.9 |
+| Situational  | event_value              | 426   |  3.9 |
+| Situational  | event_pp_change          | 236   |  2.2 |
+| Situational  | event_cumulative_return  | 108   |  1.0 |
+
+**Per-year size**: 400 questions for full years 2000–2025; 338 for 1999 (data
+limited); 130 for 2026 (partial year, only forecasts whose target is in the past).
+
+**Answer letter balance**: A 24.8% / B 25.3% / C 24.7% / D 25.1% (uniformly distributed).
+
+## Question Schema
+
+| Field           | Type            | Description                                              |
+|-----------------|-----------------|----------------------------------------------------------|
+| `qid`           | string          | Unique question ID                                       |
+| `forecastType`  | string          | `Recurrent` or `Situational`                             |
+| `subtype`       | string          | One of 8 fine-grained subtypes (see table above)         |
+| `indicator`     | string          | FRED series ID (e.g. `CPIAUCSL`, `DGS10`, `NASDAQCOM`)   |
+| `transform`     | string          | `level`, `yoy_pct`, or `yoy_pp`                          |
+| `target_period` | string          | The period the question asks about                       |
+| `info_cutoff`   | string (date)   | Latest information allowed (YYYY-MM-DD)                  |
+| `question`      | string          | Question text                                            |
+| `options`       | list of strings | 4 options, format `"A) ..."`                             |
+| `answer_letter` | string          | `A`, `B`, `C`, or `D`                                    |
+| `answer_raw`    | string          | The displayed value of the correct answer                |
+| `unit`          | string          | `%`, `pp`, `index`, `usd_billion`, `count`, `fx_rate`, etc. |
+| `year`          | int             | Convenience field for filtering                          |
+
+## Question Type Examples
+
+### Recurrent (periodic forecast)
+> *Forecast the value of [US CPI YoY Inflation Rate] for June 2010. (Information available up to 2010-04-15.)*
+>
+> A) 1.45% &nbsp;&nbsp; B) 2.04% &nbsp;&nbsp; C) 2.42% &nbsp;&nbsp; D) 1.85%
+
+### Situational — trend
+> *What was the OVERALL TREND of [US 10Y Treasury Yield] from 2010-01-01 to 2010-12-31?*
+>
+> A) Volatile &nbsp;&nbsp; B) Rising significantly &nbsp;&nbsp; C) Roughly flat &nbsp;&nbsp; D) Falling significantly
+
+### Situational — window stat
+> *During 2010, what was the LOWEST value of [VIX]?*
+
+### Situational — event-driven
+> *Following Lehman Brothers bankruptcy on 2008-09-15, what was the cumulative %
+> change in [NASDAQ] over the 20 calendar days after the event?*
+>
+> A) -14.54% &nbsp;&nbsp; B) -8.32% &nbsp;&nbsp; C) -22.71% &nbsp;&nbsp; D) -2.10%
+
+## Quality Audit (all checks passed)
+
+| Check                                              | Result |
+|----------------------------------------------------|--------|
+| Distractor uniqueness (4 distinct options)         | 0 issues |
+| Date arithmetic (`info_cutoff ≤ target_period`)    | 0 issues |
+| Required-field preservation in question text       | 0 issues |
+| Per-subtype A/B/C/D letter balance (within ±8%)    | 0 issues |
+| Event-response sanity (Lehman, COVID, etc.)        | All consistent with history |
+
+## Coverage Notes
+
+- **1999** has only 338 questions because 4 indicators
+  (`PPIFIS`, `SP500`, `DJIA`, `DTWEXBGS`) start later than 1999 on FRED.
+- **2026** is a partial year (data through April 2026); contains only forecast-style
+  questions whose target lies in the past.
+- Some indicators (`SP500`, `DJIA`) have FRED data only since 2016 (10-year limit on
+  the free tier). For historical equity exposure, `NASDAQCOM` has full 1999+ coverage.
+
+## Curated Events (Situational - event_driven)
+
+35 hand-verified events 1999–2024, including 32 FOMC impactful decisions and
+3 NBER recession periods. Examples:
+- 2001-01-03 — Fed emergency cut (dot-com)
+- 2008-09-15 — Lehman Brothers bankruptcy
+- 2008-12-16 — Fed cuts to zero (ZIRP)
+- 2010-05-06 — Flash Crash
+- 2013-05-22 — Taper Tantrum
+- 2020-03-15 — Fed emergency rate cut + $700B QE (COVID)
+- 2022-03-16 — Fed first hike of cycle
+- 2023-03-10 — Silicon Valley Bank collapse
+- 2024-09-18 — Fed first cut of cycle
+
+## Data Sources
+
+All data fetched from public APIs/sites:
+- **FRED** (Federal Reserve Bank of St. Louis) — all 49 indicator time series
+- **Federal Reserve** — FOMC meeting dates
+- **NBER** — Business Cycle Dating Committee recession dates
+- **Public news archives** — for shock-event dates and descriptions
+
+## License
+
+CC-BY-4.0. Original FRED data is in the public domain.
+
+## Related
+
+- **Original FinDeepForecast** (live benchmark): https://openfinarena.com/fin-deep-forecast/
+- **Code & dataset generation**: see the `competitions_historical/` directory in
+  the [OpenFinArena GitHub repo](https://github.com/The-FinAI/OpenFinArena).
+
+## Citation
+
+If you use this dataset, please cite the original FinDeepForecast paper:
+
+```bibtex
+@article{findeepforecast,
+  title={FinDeepForecast: A Live Benchmark for Financial Forecasting with LLMs},
+  author={OpenFinArena},
+  year={2026},
+  url={https://openfinarena.com/fin-deep-forecast/}
+}
+```