Économétrie & Séries Temporelles — Master 1 TIDE, Université Paris 1 Panthéon-Sorbonne Academic year 2024–2025
| Q1 — What? | Apply Critical Slowing Down (CSD) theory to detect early warning signals before major crypto market crashes |
| Q2 — Why? | As a system approaches a tipping point, it loses resilience — measurable through rising autocorrelation and variance |
| Q3 — How? | Rolling-window resilience indicators (AR(1), std. dev., skewness) on 500-day pre-crash windows + Kendall's Tau trend testing |
Assets & Crashes studied: Bitcoin (BTC) and Ethereum (ETH) — three major identified crashes Data: Daily closing prices over 500-day windows prior to each crash event
| Step | Description |
|---|---|
| 1. Detrending | Gaussian kernel smoothing to remove long-run trend from price series |
| 2. Indicators | Rolling-window computation of std. deviation, AR(1) coefficient, and skewness via SAS macros / PROC EXPAND |
| 3. Trend testing | Kendall's Tau applied to each indicator series to measure monotonic increase toward crash |
| Indicator | Crash (2017) | Crash (2021) | Crash (2022) | Interpretation |
|---|---|---|---|---|
| Std. Deviation | ↑ significant | ↑ significant | ↑ significant | Volatility amplification |
| AR(1) coefficient | τ = 0.72 | ↑ significant | ↑ significant | Memory increase near tipping point |
| Skewness | ↑ moderate | ↑ moderate | ↑ moderate | Distributional asymmetry |
All three crashes showed a statistically significant increase in resilience indicators, validating the CSD hypothesis on cryptocurrency markets.
analyse-crypto-csd/
├── script/
│ └── analyse_csd_crypto.sas # Full SAS pipeline (detrending, rolling indicators, Kendall's Tau)
├── data/
│ └── bitcoin.csv # BTC daily closing prices
├── report/
│ └── Memoire_critical_slowing_down.pdf # Full academic report
└── README.md
Alexis Mattei — Master 2 TIDE, Université Paris 1 Panthéon-Sorbonne
