run_po_job

Overview

run_po_job builds portfolio weights with skfolio.

Default: data_extractor_*.json — no ML needed
With forecasts: ml_engine_*.json / nn_engine_*.json — use predicted returns as μ (mu_mode: custom)

Ask for a method (equal weight, HRP, max Sharpe, DCC, …); skills fill the rest.

Parameters

string

required

Blob storage URL pointing to one of:

data_extractor_*.json — output of run_data_extraction (default path)
ml_engine_*.json — output of run_ml_job
nn_engine_*.json — output of run_dl_job

object

required

Portfolio optimization configuration.

Show config fields

object

required

Method and estimation knobs.

Show Optimization params fields

string

required

Optimization method (case-insensitive). Default when omitted: "maxsharpe".No-covariance (no covariance_mode / mu_mode knobs):

Value	Strategy
`"equal_weighted"`	Equal weight
`"inverse_volatility"`	Inverse volatility
`"random"`	Random portfolio

Covariance-based (support covariance_mode + mu_mode):

Value	Strategy
`"maxsharpe"`	Mean-Risk — maximize Sharpe
`"minvol"` / `"minimize_volatility"`	Mean-Risk — minimize volatility
`"hrp"`	Hierarchical Risk Parity
`"herc"`	Hierarchical Equal Risk Contribution
`"risk_budgeting"`	Risk Budgeting
`"max_diversification"`	Maximum Diversification
`"distributionally_robust_cvar"` / `"robust_cvar"`	Distributionally Robust CVaR

DCC aliases (auto-set covariance_mode to "dcc"):"dcc_maxsharpe", "dcc_minvol", "dcc_hrp", "dcc_herc", "dcc_risk_budgeting", "dcc_max_diversification", "dcc_robust_cvar"

Retired aliases — do not use: meanvariance, mean_variance, sharpe, maximize_sharpe, dcc_sharpe, MeanVariance, markowitz. Use "maxsharpe" / "dcc_maxsharpe" instead.

string

Covariance estimator for covariance-based methods. Default: "empirical". Values: "empirical", "ew", "ledoit_wolf", "oas", "denoise", "detone", "gerber", "shrunk", "graphical_lasso", "dcc". DCC aliases force "dcc".

string

Expected-return estimator: "empirical", "ew", or "custom" (ML/DL predictions as μ). Auto-default: "empirical" on data_extraction input; "custom" on ml/nn engine input. On data_extraction, "custom" is overridden to "empirical" (no ML predictions available).

number

Per-asset lower bound, within [-1, 1]. Default: 0. A negative value allows shorts — -0.2 permits shorting up to 20% per name. The 0 default means long-only by default, not that shorts are unsupported.

number

Per-asset upper bound, within [-1, 1]. Default: 1. Leverage is not supported: do not set it above 1.

integer

Rolling covariance window in trading days. Default: 40. It is also the warmup when you do not set an explicit split: the first n days estimate covariance and everything after them is evaluated. For DCC, estimation still uses the full train sample; the window mainly controls that warmup split.

integer

Rebalance every N trading days. Default: 1 (daily). Example: 5 ≈ weekly, 21 ≈ monthly.

object

Optional, next to Optimization params.

Show Backtest params fields

string

Explicit train/test split date, YYYY-MM-DD. Omit it and the split is derived automatically from covariance_window.

string

Caps the backtest end date, YYYY-MM-DD. Omit it to use all available data.

Dates normally come from the data you already extracted: name the sample start and end there, and let the first covariance_window days serve as warmup. Set Backtest params only when you want a specific split or a shorter evaluation window than the data allows.

Returns

{
  "status": "Succeeded",
  "output_url": "https://stqsnpprod.blob.core.windows.net/data/portfolio_optimization_35c1b5239d23.json",
  "output_name": "portfolio_optimization_35c1b5239d23.json",
  "execution_name": "po-job-abc123xyz",
  "effective_params": {
    "method": "hierarchical_risk_parity",
    "min_weights": 0.0,
    "max_weights": 1.0,
    "covariance_mode": "empirical",
    "covariance_window": 40,
    "mu_mode": "empirical",
    "rebalance_frequency": 1
  },
  "coercions": [],
  "warnings": [],
  "fallback_used": false,
  "fallback_steps": 0,
  "fallback_days": []
}

Field	Description
`status`	Job terminal status (`Succeeded`)
`output_url`	Full HTTPS URL to the output blob — pass to `run_trading_job`
`output_name`	Blob filename
`execution_name`	Job execution ID for audit/debugging
`effective_params`	The settings the run actually used, after defaults and any remapping
`coercions`	Values the job changed for you, such as `mu_mode` custom → empirical
`warnings`	Anything else worth knowing about the run
`fallback_used`, `fallback_steps`, `fallback_days`	Whether some rebalance dates fell back to equal weights, and which

A run stays Succeeded when knobs are remapped or some dates fall back to equal weights. Read effective_params, coercions and warnings before treating the result as the portfolio you asked for.

The blob contains Weights, Prices, and Meta (includes fallback flags if optimization falls back to equal weights).

Example — default path (data → PO)

{
  "input_url": "https://stqsnpprod.blob.core.windows.net/data/data_extractor_2b510101b9b7.json",
  "config": {
    "Optimization params": {
      "method": "equal_weighted",
      "rebalance_frequency": 21
    }
  }
}

Example — HRP with covariance mode

{
  "input_url": "https://stqsnpprod.blob.core.windows.net/data/data_extractor_2b510101b9b7.json",
  "config": {
    "Optimization params": {
      "method": "hrp",
      "covariance_mode": "empirical",
      "mu_mode": "empirical",
      "covariance_window": 40,
      "rebalance_frequency": 5
    }
  }
}

Example — ML-enhanced path

{
  "input_url": "https://stqsnpprod.blob.core.windows.net/data/ml_engine_7d4e2a91c0bf.json",
  "config": {
    "Optimization params": {
      "method": "maxsharpe",
      "mu_mode": "custom"
    }
  }
}

Next Step

Pass output_url to run_trading_job as input_url. Optional: run_plot_job, run_st_job, run_risk_job.

Overview

How To

Tools description

Overview

Parameters

Returns

Example — default path (data → PO)

Example — HRP with covariance mode

Example — ML-enhanced path

Next Step

​Overview

​Parameters

​Returns

​Example — default path (data → PO)

​Example — HRP with covariance mode

​Example — ML-enhanced path

​Next Step

Overview

Parameters

Returns

Example — default path (data → PO)

Example — HRP with covariance mode

Example — ML-enhanced path

Next Step