HypeX — Comics Hype Market Analytics

An end-to-end analytics project that treats "comic-book hype" as a financial market. 100 Korean manhwa/webtoon titles → a daily demand signal → a price index → returns, risk, factor backtesting, and a BI dashboard. Python · R · SQL-style modeling · PowerBI/Tableau.

⚠️ The "price" is a modeled demand/popularity index (P₀ = 100), not a real market price. Two data tracks: synthetic (a documented model, all 100 titles, 2 yrs daily) and real (actual Google Trends search interest, top 30 titles).

Skills demonstrated

Area	What
Data engineering	API ingestion (AniList GraphQL, Google Trends), entity resolution & query disambiguation, star-schema modeling, idempotent/resumable pulls with rate-limit handling
Python analytics	pandas/numpy EDA — log-returns, annualized volatility, Sharpe, drawdown, correlation, cohort analysis
Quant	cross-sectional hype-factor backtest (momentum vs reversal), strict anti-lookahead, long-short construction, turnover/drawdown
R / statistics	ARIMA forecasting (AIC grid-search), GARCH(1,1) fit by MLE from scratch, Engle-Granger cointegration — all base R
BI / visualization	interactive Plotly dashboard + a PowerBI/Tableau build guide with DAX measures

Headline results

Equal-weight market index — synthetic vs real	Hype factor — reversal beats momentum
Risk vs return across 100 titles	Return correlation (≈0 mean → real dispersion)

• The real (Google Trends) market index cools to ~94 over two years; the synthetic market centers on 100.
• Median annualized volatility ≈ 46%; mean cross-title return correlation ≈ 0 (a genuinely diversified cross-section).
• Hype mean-reverts: a contrarian (reversal) long-short earns Sharpe +2.2 while the momentum version loses by the same margin — a real regime read. (Magnitudes are inflated by synthetic mechanics; the value is the methodology.)
• R: ARIMA(2,1,2) index forecast; GARCH(1,1) with 0.92 volatility persistence; a cointegrated pair flagged for a pairs trade (ADF −4.46).

▶ Interactive dashboard: open dashboard/hypex_dashboard.html · Notebooks: EDA, backtest · Full write-up: docs/CASE_STUDY.md

Pipeline

AniList GraphQL ─┐
                 ├─► universe (100 titles, themed/dated cohorts)
Google Trends ───┘            │
                              ▼
        demand signal ──► price model (z-score → hype index → smoothing → price)
        (synthetic + real)            │
                                      ▼
                 Python EDA + quant backtest · R stats · Plotly/PowerBI dashboard

Reproduce

pip install -r analytics/requirements.txt          # pandas, plotly, seaborn, statsmodels, jupytext, pytrends
python analytics/universe/fetch_anilist.py         # 1. universe from AniList
python analytics/pipeline/generate_synthetic.py    # 2. Track 1 (synthetic, 100 titles)
python analytics/trends/build_query_map.py         # 3. Track 2 query map ...
python analytics/trends/fetch_trends.py            #    ... pull Google Trends (top 30)
python analytics/pipeline/reprice_real.py          #    ... price the real signal (weekly)
python analytics/pipeline/combine_tracks.py        # 4. unified dataset
jupytext --to notebook --execute analytics/notebooks/01_eda.py
jupytext --to notebook --execute analytics/notebooks/02_backtest.py
Rscript analytics/r/report.R                       # 5. R statistical report
python dashboard/build_dashboard.py                # 6. interactive dashboard

Analytics layout

analytics/
  universe/   fetch_anilist.py · titles.csv · candidates.csv     # API ingestion + curation
  trends/     build_query_map.py · fetch_trends.py · query_map.csv
  pipeline/   pricing.py · generate_synthetic.py · reprice_real.py · combine_tracks.py
  notebooks/  01_eda.ipynb · 02_backtest.ipynb
  r/          report.R
  reports/    figures/*.png · *_findings.md
  data/exports/  prices_all.csv · titles_dim.csv · market_index.csv · metrics*.csv
dashboard/    build_dashboard.py · hypex_dashboard.html · POWERBI_TABLEAU_GUIDE.md

The data platform underneath

The analytics sit on the original HypeX application — a FastAPI backend, Next.js frontend, and PostgreSQL data platform with a plugin-based ingestion system and the same documented price model. See docs/ARCHITECTURE.md and docs/DATA_MODEL.md. The analytics track above is deliberately self-contained (CSV in/out, no database required) so it reproduces anywhere.

Honest data note

Title metadata and real search interest are real (AniList, Google Trends). Prices are a modeled index, not traded values, and the synthetic track is clearly labeled as such. Sweet Home's real signal is flagged as noisy (a large Netflix adaptation inflates search). This transparency is the point — the project demonstrates the analytics pipeline, not a claim to predict comic popularity.

License

Released under the MIT License.