Skip to content
Northstar Capital

Northstar Capital · Systematic trading

A fixed point in moving markets.

Northstar Capital is a systematic trading firm. We take trading strategies from published research, run them as production software, and reconcile live results against simulation every day. Our field notes explain the methods behind that work, using public data.

Read the field notesAbout the firm

Our approach

Research

We study sources of return documented in the academic literature and test whether they still hold once realistic trading costs and delays are charged. Try enough variations of any idea and one will look good by chance, so we record every variation we test and demand more of the best one the more attempts it took to find. We also set aside a stretch of market history and do not look at it until a strategy’s rules are final; it then serves as a last, independent test. What survives is far less likely to be a statistical accident.

Engineering

We write the software that runs our strategies and hold it to the standards of any system that handles money. No change reaches live trading until automated tests have passed, and every version we run is recorded, so any result, in testing or in the market, can be traced to the exact code behind it. While a strategy runs, the system keeps checking the conditions it depends on, such as up-to-date market data and positions within their limits. If one fails, trading stops before a fault can become a costly trade.

Operations

We run our strategies in live markets and check every trading day that they behave as the research said they would. The positions and returns on our broker statements are compared with a simulation of the same day, and any gap is traced to its cause, such as late data, a timing difference or a trade executed at a worse price than expected. Large orders are spread over time so that our own trading does not push prices against us, and risk controls that sit above every strategy can cut exposure when markets turn.

Public data · since 1926

100 years of the US market

One unit invested in the whole US stock market in July 1926, dividends reinvested, on a log scale. The shading is the distance below the previous high. Hover over the chart, or move along it with the arrow keys, to read any week; open a decline’s label for its dates and the field notes that study it.

Latest31 August 2026Growth of 118,129Below the previous high−1.5%

US stock market total return, 1 July 1926 to 31 August 2026, log scale: one unit grew to 18,129. Drawdowns from the previous high are shaded; the deepest was 84%, from 1929 to 1932. Use the arrow keys to move through the weeks, Shift with an arrow for a year, Page Up and Page Down for a decade.

1101001k10k1930194019501960197019801990200020102020−50%18,129Below the previous high
1929The 1929 crash and the Depression−84.1% from 3 September 1929 to 8 July 1932Back to the 1929 high on 6 February 1945, 15.4 years after the peak.The fall that began in September 1929 lasted 2.8 years. Even with dividends reinvested, the market regained its 1929 high only in February 1945.How volatility targeting reshaped this drawdown →The 200-day rule in the deepest declines →
1937The 1937–38 recession−51.0% from 10 March 1937 to 31 March 1938Back to the 1937 high on 28 June 1943, 6.3 years after the peak.A second bear market, which began while the market was still below its 1929 high: at the trough it stood 60% below that high.The 200-day rule, decade by decade →
1973–74The 1973–74 bear market−48.2% from 11 January 1973 to 3 October 1974Back to the 1973 high on 28 December 1976, 4.0 years after the peak.The oil embargo, rising inflation and recession: a grinding decline over 1.7 years rather than a crash.The 200-day rule in the deepest declines →
1987Black Monday−33.1% from 25 August 1987 to 4 December 1987Back to the 1987 high on 15 May 1989, 1.7 years after the peak.On 19 October 1987 the market fell 17.4%, the largest one-day fall in the record. The decline bottomed on 4 December 1987.Why lower volatility is not a lower worst case →
2000–02The dot-com bust−49.2% from 24 March 2000 to 9 October 2002Back to the 2000 high on 23 October 2006, 6.6 years after the peak.The collapse of technology valuations, then recession: 2.5 years from the peak to the trough.The 200-day rule in the deepest declines →
2008The global financial crisis−54.6% from 9 October 2007 to 9 March 2009Back to the 2007 high on 28 February 2012, 4.4 years after the peak.The deepest decline since the 1930s, driven by the collapse of the housing and credit markets.The 200-day rule through 2008, step by step →
2020The pandemic crash−34.2% from 19 February 2020 to 23 March 2020Back to the 2020 high on 5 August 2020, 6 months after the peak.A fall of 34% in 23 trading sessions, regained 4 months after the trough.What volatility targeting cannot see coming →
2022Inflation and rising rates−25.6% from 8 November 2021 to 14 October 2022Back to the 2021 high on 19 December 2023, 2.1 years after the peak.The broad market peaked in November 2021. Its closing lows of 30 September, 12 October and 14 October 2022 were within a tenth of a percentage point of one another.Why gradual declines pass volatility rules by →
Kenneth R. French Data Library (Dartmouth): the daily total return of the CRSP value-weighted US market, 1 July 1926 to 31 August 2026, in US dollars, before inflation, costs and taxes. Drawn from the last session of each week and each named decline’s high and low; declines are measured on daily closes. Public market data, not the results of any account; past returns do not predict future ones.

Knowledge Centre

Field notes

A technical curriculum for engineers, analysts and students who want to research and run systematic strategies, from research foundations to live operations. Figures built from data come with the code that reproduces them.

Start hereIntroduction to the series →Who the notes are for, how they are organised, and where to start if you arrive with a specific problem.FoundationsMarket anomaliesMost published return anomalies shrink once they are known, and the reason an anomaly exists predicts how it will fail.Introductory · 21 min readFoundationsData for systematic tradingThe data errors that make a backtest look better than it is, and how to catch each one in code before a strategy reads it.Intermediate · 28 min readFoundationsIndicators and the 200-day moving averageOn a century of daily US data, the 200-day moving average reliably lowers risk, while its return advantage depends on the decade and on the execution.Intermediate · 19 min readResearch infrastructureQuantConnect and LEANOne open-source engine runs the same algorithm in backtest and live, and its defaults decide how far the two drift apart.Introductory · 17 min readResearch infrastructureRunning backtests at scaleA backtest can be trusted only if it can be reproduced and counted, and both have to be built into the pipeline before the first sweep runs.Intermediate · 36 min readPortfolio construction and riskVolatility targetingSizing positions by forecast risk to keep the risk carried roughly constant, and where the rule stops protecting.Intermediate · 27 min readPortfolio construction and riskBuilding strategies from componentsSelf-contained components with their own signal, sizing and state, tested alone and in the book, and why a component’s standalone Sharpe ratio says little about what it adds.Intermediate · 20 min readPortfolio construction and riskCombining strategies into one bookCorrelation depends on how often it is measured, and so does every decision about how to size, net and cap strategies in one book.Intermediate · 22 min readExecutionTWAP, slicing and jitterA trade’s cost is measured against the price when it was decided; splitting an order keeps that cost down, and an irregular schedule makes it harder for others to trade ahead of it.Intermediate · 20 min readProduction operationsGoing liveGoing live tests the machinery around a strategy, and most of what can go wrong there can be measured before money moves.Intermediate · 19 min readProduction operationsMonitoring a live bookEnd-of-day checks that show their evidence, and that say so when they could not look.Intermediate · 14 min readProduction operationsAlerting that people act onAn alert rule is a classifier: score it on what it catches, what it misses, how late it speaks and how often it cries wolf.Intermediate · 15 min readProduction operationsLive versus backtest reconciliationEvery gap between a live book and its backtest has a cause, and a daily comparison of holdings, decisions and returns names it.Intermediate · 18 min read

Principles

What we hold to

  1. 01

    Decide before we look

    Before the first test, we fix the rules and write down the result that would make us drop the idea, so the data cannot quietly shape the conclusion.

  2. 02

    Count everything

    We record every backtest, failures included, because the best of many attempts tends to look better than it really is.

  3. 03

    Size for the bad day

    We size positions for what can plausibly go wrong, such as overnight gaps and losses that arrive together, and not for a typical day.

  4. 04

    Prefer robust to clever

    We prefer simple rules that work across many markets and settings. A rule tuned to win one backtest usually disappoints in the next.