This is a project built as a study for Saarland University's program “Data and Society”. We argue that digital media piracy is a predictable market correction. We hypothesize piracy emerges when market failures, specifically price disparity, fragmentation, and poor preservation fail consumers.

🔑 Key findings

1. Fragmentation, not poverty, tracks the piracy surge. As the global streaming market fragmented from 2018 to 2023 (HHI fell 60%, from 5,276 to 2,101 on Parrot Analytics demand-share data), actual piracy visits rose 10.6% to an all-time high of 141 billion (MUSO). Meanwhile Google searches for piracy declined; we call this the Google Trends paradox. Piracy stopped being something people search for and became infrastructure (Stremio, IPTV boxes, Telegram channels), which is exactly what a maturing shadow market looks like.

2. Wealth does not predict piracy. Across 106 countries, estimated hourly wages explain about 1% of the variance in piracy search volume (OLS R² = 0.013; Spearman ρ = −0.001, p = 0.99). This falsification test rules out the "piracy is just poverty" explanation and strengthens the fragmentation story above.

3. Delistings trigger archival piracy. An event study across ~70 delisted games shows piracy search interest peaking within roughly ±15 days of the delisting event (month-level date granularity), consistent with piracy acting as a preservation mechanism for media that is going extinct.

The takeaway: piracy behaves like a barometer for market failure. The industry beats it by competing on service (centralized, frictionless, permanent access), not by enforcement.

👥 Team

Built by Team 5 for the Data and Society seminar: Akshay Ashok, Ali Haider Khan and Vishakh Kantharaj.

• Akshay Ashok: fragmentation barrier (Parameter 2), HHI computation on Parrot Analytics and Sandvine data, MUSO alignment, Google Trends collection via pytrends, report integration
• Ali Haider Khan: economic barrier (Parameter 1), wage and income data engineering, correlation analysis
• Vishakh Kantharaj: original Shadow Correction concept and preservation barrier (Parameter 3), delisting event study

🐍 Dependencies

One command gets you everything:

pip install -r requirements.txt

Or piece by piece, if that's how you like living:

Pandas: You know what pandas is for

pip install pandas

Numpy: math and whatnot

pip install numpy

MatplotLib: graphs and whatnot

pip install matplotlib

Seaborn: Used to create and vizualize graphical data

pip install seaborn

Pytrends: Unofficial API to query google trends

pip install pytrends

SciPy: Fundamental algorithms for scientific computing in Python

pip install scipy

StatsModels: Statistical computations and models for Python

pip install statsmodels

📂 Structure

The repo contains 3 main folders each necessary to be able to reproduce the study locally please keep the names of the folders unchanged.

• Data-sets: contains all the extracted datasets for the study

• Scripts: contains the necessary python files to be able to generate some of the data required for the final analysis

• Results: contains the result generated during our self-analysis (Final report based on these results)

• assets: plots exported from the report notebook for this README

💻 Setup

Follow these steps to setup and run the tests on your local device.

1. Read this file ( Du bist Smart :O )

2. download the latest release of python and jupyter notebook from thier official pages (links above)

3. download the latest release this should give you all the necessary files

4. Install all the dependencies mentioned above using the pip command

5. Proceed to "Scripts" folder (This will be your root folder to run some of the data gen)

6. Follow the instructions in "scripts-readme.txt" to run the data generation process

P.S if you just want to see the final results and project analysis instead run the "project_report.ipynb" file

⚠️ Honest limitations

The global fragmentation correlation runs on n = 6 years, so the directional trend (HHI −60%, piracy +10.6%) is clear but the correlation itself is not statistically significant. Sandvine changed measurement methodology in 2017, COVID (2020) is an anomaly we handle via a robustness check, and delisting dates only have month-level precision. Full discussion in section 2.8 of the report notebook.