CyclicalResearch
Proprietary cyclical analysis system for crypto markets that blends Digital Signal Processing, cycle theory and machine learning. It identifies dominant cycles across multiple time horizons, recognizes the current market regime and projects forward scenarios with confidence bands. A pattern-recognition engine matches the live cycle against the market's historical memory to estimate its most likely evolution. All inside an interactive dashboard that makes the cyclical map readable and navigable in real time. Stack: Python, Polars, NumPy/Numba, SciPy, Dash.
The problem
Crypto market cyclicality is visible by eye but hard to measure: classic indicators lag and can't tell a cycle that continues from a regime that has changed. What was needed was reading the dominant cycles across several horizons at once, and knowing which phase you're in as it happens — not in hindsight, once the move is over.
How we tackled it
The system treats price history for what it is, a signal: DSP filters isolate the cyclical components, spectral analysis finds the dominant periods, and a regime classifier says which state the market is in. A pattern-recognition engine compares the current cycle against the market's historical memory and projects scenarios with confidence bands, rather than a single number implying false precision.
Technical choices
Python with Polars for columnar processing and Numba for JIT-compiled maths: multi-timeframe analysis has to stay interactive, not become an overnight batch. SciPy for the spectral work, Dash and Plotly for the dashboard, CCXT for market data. One non-negotiable constraint: no lookahead — every value uses only the information available at that bar.
What it includes
Dominant-cycle extraction across timeframes, spectral analysis, market regime classification, a pattern-recognition engine over historical memory, scenario projection with confidence bands, and an interactive dashboard for navigating the cycle map. Automated market data ingestion via CCXT, with a reproducible pipeline: same data and parameters, same result.