Modules / Finance

Version 0.1.0 · Apache-2.0 · Python

Holoscan Finance

Holoscan Finance builds a streaming financial-services application as a Holoscan graph. Replay a validated sequence of market ticks, keep a rolling window of statistics for each symbol, and mark a portfolio against its limits. Prices and positions use exact decimals, and the reference application runs offline on checked-in synthetic data.

Technical reference

This module is a technical reference. It is not investment, trading, accounting, regulatory, or risk-management advice, and it is not a production risk engine.

At a glance

Package	holoscan-finance
Version	0.1.0
Python namespace	holoscan.finance
Holoscan SDK	4.3.0 or later, tested with 4.3.0
Platforms	x86_64, aarch64
Language	Python 3.10 through 3.13
License	Apache-2.0
Maintainer	Bruce Hashemian, NVIDIA
Debian package	holoscan-finance
PyPI package	holoscan-finance

Operators

Operator	What it does
MarketDataReplayOp	Emits a finite sequence of validated market ticks, in order, without waiting on wall-clock time
RollingStatisticsOp	Keeps an independent window per symbol and emits returns, a moving average, and return deviation
PortfolioRiskOp	Marks configured long and short positions with the latest prices and evaluates portfolio limits

load_market_ticks_csv reads the module’s CSV format: a timezone-aware timestamp, a strictly increasing sequence, a symbol, a positive price, and a three-letter currency.

RollingStatisticsOp emits one features message for every tick. The moving average and return deviation stay empty until the window has enough observations. Return deviation is a descriptive window statistic. It is not annualized volatility or Value at Risk.

PortfolioRiskOp emits a snapshot after every features message, including a partial snapshot while some positions are still unpriced. Limits are checked only when every position has a price. A breach is data on the snapshot, not an error that stops the graph.

Data models

Model	What it carries
MarketTick	One observation: timestamp, sequence, symbol, price, and currency
MarketFeatures	That tick plus the symbol’s current rolling window
Position	A signed quantity and average entry price. Positive is long, negative is short.
RiskLimits	Optional caps on gross exposure, absolute net exposure, and unrealized loss
PortfolioSnapshot	Marks, exposures, unrealized P&L, unpriced symbols, and any limit breaches

Prices, quantities, and P&L use decimal.Decimal. Construct them from strings.

Sample application

intraday_risk_pipeline replays synthetic ticks, computes a short rolling window, marks a small long and short portfolio, and logs each snapshot and limit breach. It needs no network access and no market-data service.

./holohub run intraday_risk_pipeline

Use it in an application

from decimal import Decimal

from holoscan.core import Application
from holoscan.finance import (
    MarketDataReplayOp,
    PortfolioRiskOp,
    Position,
    RiskLimits,
    RollingStatisticsOp,
    load_market_ticks_csv,
)

class RiskApplication(Application):
    def compose(self) -> None:
        ticks = load_market_ticks_csv("market_ticks.csv")
        replay = MarketDataReplayOp(self, ticks=ticks, name="market_data_replay")
        statistics = RollingStatisticsOp(self, window_size=20, min_periods=5, name="rolling_statistics")
        risk = PortfolioRiskOp(
            self,
            positions=(Position(symbol="NVDA", quantity=Decimal("100"), average_price=Decimal("150.00")),),
            limits=RiskLimits(max_gross_exposure=Decimal("25000"), max_abs_net_exposure=Decimal("25000"), max_unrealized_loss=Decimal("1000")),
            name="portfolio_risk",
        )
        self.add_flow(replay, statistics, {("tick", "tick")})
        self.add_flow(statistics, risk, {("features", "features")})

RiskApplication().run()

Install the Holoscan SDK first, then pip install holoscan-finance.

Source is in the Holoscan Finance repository.

All modules →