Skip to content

benchmarks.md

Latest commit

 

History

History
56 lines (43 loc) · 2.54 KB

benchmarks.md

File metadata and controls

56 lines (43 loc) · 2.54 KB

Performance Benchmarks

These are the measured timings for MeridianAlgo. They come from the hand run recorded in TEST_RESULTS.md. No numbers here are estimated. They are wall clock times from a single machine, so treat them as a guide to relative cost rather than a hardware specification.

Test stack

All timings were taken on the following stack against a 6 asset, 1000 day synthetic price and return set. The synthetic data needs no network access and keeps the run reproducible.

  • Python 3.14.5
  • numpy 2.4.6
  • pandas 3.0.3
  • scipy 1.17.1

This stack is newer than the versions the package was written against, so the run also serves as a forward compatibility check.

Core operations

Operation Detail Time
Mean variance max sharpe 6 assets 3.8 ms
Hierarchical risk parity 6 assets 12.3 ms
Risk parity 6 assets 3.9 ms
Black Litterman 6 assets 0.8 ms
Black Scholes price and greeks single option 0.5 ms
Implied volatility solve single option 2.1 ms
Merton model calibration single firm 1.7 ms
CDS fair spread single name 0.3 ms
GBM simulation 10k paths, 252 steps 53.6 ms
Heston simulation 5k paths, 252 steps 40.7 ms
Jump diffusion simulation 5k paths, 252 steps 52.2 ms
GARCH fit 1000 observations 77.2 ms
Realized volatility, five estimators 1000 days 4.1 ms
CPPI run 1000 days 4.4 ms
Full performance metrics 1000 days 3.6 ms

Headline figures from the README

The README performance table reports a handful of the same operations, with a larger GBM path count.

Operation Detail Time
Mean variance max sharpe 6 assets 3.8 ms
Hierarchical risk parity 6 assets 12.3 ms
Black Scholes price and greeks single option 0.5 ms
Merton model calibration single firm 1.7 ms
GBM simulation 100k paths, 252 steps around 0.3 s
Heston simulation 5k paths, 252 steps 40.7 ms
GARCH fit 1000 observations 77.2 ms
Realized volatility, five estimators 1000 days 4.1 ms
Full performance metrics 1000 days 3.6 ms

Notes

The numerical core is fast and the results come back as typed result objects and pandas structures. Optimization, risk, credit, volatility, Monte Carlo, derivatives, and analytics all returned sensible numbers in the run. The GARCH fit timing applies to the maximum likelihood path, which needs the volatility extra.

For the full methodology and the complete pass and fail breakdown see TEST_RESULTS.md.