Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Simulated daily returns · mkurman | DecimalAI

mkurman/pymc-markets

SkillScore70Maintained

Bayesian inference for financial markets using PyMC. Stochastic volatility models, regime-switching, Bayesian portfolio optimization, factor models, and Markov chain Monte Carlo for risk estimation.

financeby mkurman

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

Finance & Accounting pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/mkurman-pymc-markets/SKILL.md

decimalai skills pull mkurman-pymc-markets --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/mkurman-pymc-markets/SKILL.md https://app.decimal.ai/s/mkurman-pymc-markets/SKILL.md

Raw file: /s/mkurman-pymc-markets/SKILL.md ↗ · machine-readable summary at /s/mkurman-pymc-markets.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

23 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 8d ago

Yes — +9% lift-21% tokens0% turns70% of whole cases pass with skill (16/23)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-14	✗→✓	▲ Improved	-26%	0%
case-23	✗→✓	▲ Improved	-28%	0%
case-04	✓→✓	＝ Same ✓	-64%	0%
case-07	✓→✓	＝ Same ✓	-21%	0%
case-08	✓→✓	＝ Same ✓	19%	0%

Showing 5 of 23 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 8/3/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

mkurman / zorai

skills/scientific-skills/pymc-markets/SKILL.md

View on GitHub ↗

License MITImported 8/3/2026Version v1

Skill contents

SKILL.md

Overview

PyMC provides Bayesian inference for financial modeling: stochastic volatility, regime-switching, Bayesian portfolio optimization, factor models, and MCMC risk estimation using the NUTS sampler. ArviZ provides diagnostics and visualization.

Installation

bash
uv pip install pymc arviz

Stochastic Volatility Model

python
import pymc as pm
import numpy as np
import arviz as az

# Simulated daily returns
returns = np.random.randn(500) * 0.02

with pm.Model() as sv_model:
    sigma = pm.InverseGamma("sigma", alpha=2, beta=1)
    log_vol = pm.GaussianRandomWalk("log_vol", sigma=sigma, shape=len(returns))
    obs = pm.Normal("returns", mu=0, sigma=pm.math.exp(log_vol / 2), observed=returns)
    trace = pm.sample(1000, tune=1000, chains=4)

print(az.summary(trace, var_names=["sigma"]))
az.plot_trace(trace)

References

Want to see if mkurman/pymc-markets would activate on your phrasing? Try it before you install — no LLM call, no signup.

Latest benchmark · publisher runVerified

8/21/2026, 11:01 UTC · 23 cases

Lift

+9%

70% vs 61%

Agent turns

1.0 vs 1.0

Tokens

-22%

1,563.35 vs 1,990.83

fail → pass · 2pass → pass · 14fail → fail · 7

Per-case results (23)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-01 Can you write a Python script using PyMC to build a Bayesian stochastic volatility model for a set of daily financial returns? I need the script to define the latent log-volatility process, fit the posterior using MCMC sampling, and output diagnostic summary statistics using ArviZ.	fail→fail	32,244	13,834	-57%	1	1	0%	6,197	2,920	-53%	0	0	—
case-02 I have daily stock return data and want to model time-varying volatility using a Gaussian random walk for log-volatility in PyMC. Please provide a complete Python code snippet that constructs the probabilistic model, samples the posterior distribution, and generates trace plots for model diagnostics.	fail→fail	16,595	11,647	-30%	1	1	0%	3,238	2,730	-16%	0	0	—
case-03 Please show me how to implement a PyMC probabilistic model for estimating financial market volatility under a stochastic volatility framework. The response should include code that sets up priors for latent volatility variance, samples the chain using the default MCMC sampler, and prints parameter convergence metrics.	fail→fail	18,409	15,627	-15%	1	1	0%	3,590	3,358	-6%	0	0	—
case-04 What is the recommended terminal command to install PyMC along with ArviZ using the uv package manager?	pass→pass	7,789	1,343	-83%	1	1	0%	1,253	457	-64%	0	0	—
case-05 When configuring the standard deviation parameter sigma for a Gaussian random walk on log-volatility in PyMC, should I use an Exponential prior or an InverseGamma prior? Please provide the exact PyMC distribution call with its hyperparameter values.	fail→fail	10,375	7,607	-27%	1	1	0%	1,694	1,535	-9%	0	0	—
case-06 I am writing a stochastic volatility model script for S&P 500 returns. I want to build a Gaussian random walk state variable for log-volatility. Should I build a manual random walk via np.cumsum or use PyMC's built-in random walk primitive?	fail→fail	17,618	14,946	-15%	1	1	0%	3,084	3,018	-2%	0	0	—
case-07 In a Bayesian stochastic volatility model, how should the latent variable `log_vol` transform into the standard deviation parameter for the normal return likelihood in PyMC?	pass→pass	12,659	8,284	-35%	1	1	0%	2,351	1,864	-21%	0	0	—
case-08 What call to `pm.sample` should be used to run MCMC inference on the stochastic volatility model with 1000 draw steps, 1000 tuning steps, and 4 parallel chains?	pass→pass	3,609	2,644	-27%	1	1	0%	635	757	+19%	0	0	—
case-09 How do I generate an ArviZ statistical summary table for only the latent volatility scale parameter `sigma` after MCMC sampling?	pass→pass	11,004	3,986	-64%	1	1	0%	1,861	1,002	-46%	0	0	—
case-10 What ArviZ function call is used to plot MCMC trace chains for visual convergence checking of a PyMC model trace?	pass→pass	3,345	2,623	-22%	1	1	0%	509	693	+36%	0	0	—
case-11 Many financial tutorials use StudentT distributions to capture heavy tails, while standard stochastic volatility models handle variance shifts through log-volatility. Which likelihood distribution is specified in standard PyMC stochastic volatility models for returns given `log_vol`?	pass→pass	6,029	5,573	-8%	1	1	0%	971	1,244	+28%	0	0	—
case-12 Create a Python snippet that generates 500 simulated return observations (`returns = np.random.randn(500) * 0.02`) and fits a stochastic volatility model using PyMC context manager `with pm.Model() as sv_model:`.	pass→pass	8,716	7,239	-17%	1	1	0%	1,647	1,829	+11%	0	0	—
case-13 Write a code snippet that prints the ArviZ summary table specifically for parameter `sigma` from an MCMC posterior trace `trace`.	fail→fail	2,905	1,729	-40%	1	1	0%	448	557	+24%	0	0	—
case-14 What shape (alpha) and scale (beta) parameter values should be assigned when instantiating `pm.InverseGamma` for the log-volatility innovations in the standard PyMC stochastic volatility script?	fail→pass	14,709	9,460	-36%	1	1	0%	2,565	1,902	-26%	0	0	—
case-15 Should standard Python `math.exp` or `np.exp` or `pm.math.exp` be used inside a PyMC model context when computing the scale vector for the return likelihood from log-volatility?	pass→pass	9,440	6,531	-31%	1	1	0%	1,616	1,441	-11%	0	0	—
case-20 How do I fit a classical (frequentist) GARCH(1,1) model on S&P 500 daily return data in Python to estimate conditional variance using maximum likelihood estimation?	pass→pass	13,650	7,097	-48%	1	1	0%	2,671	1,664	-38%	0	0	—
case-16 When configuring `pm.sample()` for stochastic volatility inference, how many tuning steps are specified alongside 1000 sampling draws across 4 chains?	pass→pass	6,705	1,732	-74%	1	1	0%	1,175	608	-48%	0	0	—
case-17 Write a two-line Python code block: first printing the ArviZ parameter summary for `sigma`, and second displaying trace plots for the entire trace.	pass→pass	2,655	2,064	-22%	1	1	0%	463	632	+37%	0	0	—
case-18 How is the length of time series returns passed to the latent Gaussian random walk distribution in PyMC so that its dimension matches the data?	pass→pass	14,037	5,224	-63%	1	1	0%	2,483	1,168	-53%	0	0	—
case-19 Show the full PyMC model context definition `with pm.Model() as sv_model:` including prior, latent walk, and likelihood nodes for daily return vector `returns`.	fail→fail	12,505	7,916	-37%	1	1	0%	2,378	1,862	-22%	0	0	—
case-21 I need a Python function to calculate the European call option price and delta using the analytical Black-Scholes-Merton formula given spot price, strike, risk-free rate, time to maturity, and volatility.	pass→pass	8,296	7,345	-11%	1	1	0%	1,788	1,922	+7%	0	0	—
case-22 Provide a Python code snippet using pandas and numpy to backtest a simple moving average crossover strategy (50-day and 200-day) on historical stock prices, calculating total return and Sharpe ratio.	pass→pass	11,697	9,311	-20%	1	1	0%	2,382	2,227	-7%	0	0	—
case-23 In PyMC MCMC estimation for stochastic volatility models, how many parallel sampling chains are configured in the standard sampling call?	fail→pass	5,129	1,931	-62%	1	1	0%	790	567	-28%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted. The headline lift of +9 percentage points is the difference between those two pass rates over the 23 comparable cases.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 60.9%
Pass rate with the skill: 69.6%
Cases attempted: 23
Cases counted toward lift: 23
Cases that regressed: 0
Skill version measured: v1
Run completed: 8/21/2026, 11:01 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-14

Prompt

What shape (alpha) and scale (beta) parameter values should be assigned when instantiating `pm.InverseGamma` for the log-volatility innovations in the standard PyMC stochastic volatility script?

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signals1 of 4 complete

BenchmarkPass rate WITH skill on a verified run70

Live evalNeeds ≥20 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

Efficiency · separate axis (at equal correctness)+0% turns-20% tokens

Shown for transparency; not credited unless a verified run shows no correctness regression.

Backed by 1 of 4 signals — each added signal makes the score more robust.

Verified on gemini-3.6-flash · re-tested 8 days ago

Other measured skills in the registry, with their headline benchmark lift.