About
About Northstar Capital
We trade our own capital with systematic strategies, and we build and operate the infrastructure they run on.
The work is half statistics and half production engineering. A strategy trades real money only after its central idea has survived testing on data it was not designed on, with realistic costs charged. It keeps trading only while its live results match what the research predicted.
What we do
From hypothesis to live book
Research
We start from return sources documented in the academic literature and test whether they survive realistic costs, execution delays and constraints. Most do not. The few that do are written down in full before we look at the stretch of history set aside for the final test.
Engineering
Our strategies, research pipelines and monitoring are software we write and maintain ourselves. No change reaches the live system without passing an automated test suite, so a mistake is caught in testing and not in the market.
Operations
The code that trades is the code we tested, so what we measured in research is what runs in the market. Every day, live results are compared with a simulation of the same trading.
How we work
Live results feed back into research
A gap between live and simulated results is logged and treated as a research question. The research record keeps each backtest and decision, so the question leads straight back to the work behind the strategy.
Research process
Each rule below exists to stop a lucky backtest from reaching the live book.
- Hypotheses first
- Before a test runs, we record the effect we expect to exploit and why it should exist, the markets we will trade, the targets we will judge it by, and the result that would make us abandon the idea.
- Every backtest counted
- We record every backtest we run, including the ones that failed, so a lucky result can be told apart from a real one. The more attempts behind the best result, the higher the bar it has to clear.
- Untouched test data
- We set aside part of the history and do not look at it until the strategy’s rules are final. If we change the rules after seeing that data, it no longer counts as a fair test, and the work goes back to the start.
- Costs from the first run
- Commissions, spreads, slippage and execution delay are modelled from the first backtest. We treat a result that exists only before costs as no result.
- Robustness over precision
- We prefer settings whose results change little when nudged, and we test how results move when lookbacks, timing and inputs are varied. A result that depends on one exact setting is usually luck.
Engineering standards
The same code runs in research, in testing and in live trading.
- Reproducible builds
- Every version we run is recorded, so we can rerun any past backtest on the same code and data and get the same answer.
- Run-time safety checks
- Position limits, the freshness of market data and the size of each order are checked continuously while a strategy runs. A failed check stops trading and alerts us.
- Tests that break things on purpose
- We deliberately break the system in testing to prove that each safety check fires when it should. A check earns its place by catching the error it was written to catch.
- Automatic records
- Inputs, decisions, orders and fills are recorded as the system runs, so any trading day can be reconstructed from the record.
Live operations
Each trading day is checked and reconciled before the next session opens.
- Daily monitoring
- After each session we check what we hold, how much risk it carries, the prices we traded at and the quality of the data the strategies saw, so a problem is found before the next session opens.
- Live-versus-simulation reconciliation
- We compare live positions and returns with a simulation of the same period, using broker statements as the reference, and explain every material difference before it can grow.
- Actionable alerting
- Every alert names an action. One that fires and changes nothing is fixed or removed, because alerts that are routinely ignored teach people to ignore the one that matters.
- Risk controls above the strategies
- Risk controls sit above the strategies and act on their own measures of market conditions, independent of any strategy’s signal. They cut exposure as conditions deteriorate, and a kill switch can close every position at once.
Principles
What we hold to
- 01
Decide before we look
We write down targets, rules and the result that would make us drop an idea before the first backtest runs. A criterion chosen after seeing the data only describes that data; it cannot test anything.
- 02
Count everything
Every backtest goes into a ledger, failures included. We discount the best result for the number of attempts behind it, because anything left out of the count makes a result look better than it is.
- 03
Size for the bad day
Position sizes follow from what can plausibly go wrong, including overnight gaps, correlated losses and the slow declines that volatility measures miss. The average day says little about risk.
- 04
Prefer robust to clever
A simple rule that holds across markets, periods and nearby parameter values is worth more to us than a finely tuned one that wins a single backtest.
The Knowledge Centre
Why we publish
Most of what makes systematic trading work is method, and method can be shared without giving away a strategy. The field notes set out the public research, tests and operating habits we rely on, with the primary sources behind each claim. Writing them also disciplines our own reasoning: an argument that cannot be written down clearly is rarely understood well enough to trade.
The notes describe published methods, not the rules or parameters of our strategies. Where the evidence is mixed or an effect has faded, they say so. Start with the introduction to the series.
The name
Why Northstar
The pole star holds nearly still while the rest of the sky turns, which is why navigators steered by it. A systematic process is meant to work the same way: its rules stay fixed while markets move around them.
What this site is not
We do not manage money for others. We publish no returns, positions or strategy rules. Sign-in is for our internal tools.
Contact
Questions and corrections on the field notes are welcome; contact details are in the imprint.