Bespoke quantitative solutions
Your investment process is your own. We build quantitative research, portfolio tools and AI workflows to match.
01 / What we make
Nothing comes off the shelf. We design each tool around your data, your constraints and the decisions you need to make.
Signals and research tools tailored to your investment horizon, universe and data—from prices and filings to news and web traffic.
Tools for portfolio construction, position sizing and risk management, built around your day-to-day investment decisions.
Language models integrated into your research and trading workflows, with the controls and transparency your investment process demands.
02 / AI for equities
Filings, earnings calls and news contain more than any team can read. Put AI to work organising the evidence, so analysts can focus on what it means.
Compare disclosures and earnings calls over time. Surface changes in guidance, management language and reported risks, with links back to the source.
Extract comparable facts from company documents and organise them by issuer, sector or theme. Bring the relevant evidence into screening and peer analysis.
Prepare research briefs and flag developments for review. Separate reported facts from model interpretation, so every investment judgement rests on evidence your team can check.
03 / AI for quantitative trading
A model score is not a trading strategy. Test whether information extracted by AI adds value at your horizon, after costs and alongside the signals you already use.
Translate filings, news and transcripts into structured event, sentiment and topic features. Align each observation with when the information became available to trade.
Compare model features with simple baselines and evaluate them out of sample. Account for look-ahead bias, turnover, transaction costs and signal decay before considering live use.
Feed validated outputs into your existing portfolio and trading systems. Keep position limits, risk checks and execution permissions separate from the model.
04 / AI integration
Start with a specific research or trading task, not a mandate to use AI everywhere. Define what a useful result looks like, then build the connections and controls it needs.
Work with your licensed feeds, internal research and approved documents. Choose model and deployment options around data permissions, confidentiality and latency requirements.
Retain source references and model versions, evaluate outputs against agreed examples, and route uncertain results for human review. Set clear boundaries on what the system can do.
Run alongside the current workflow before relying on it. Assess accuracy, time saved and operating cost, with monitoring and a way to pause or roll back changes.
05 / About
9epoch is led by Nathan Szeitli, a quant who brings research rigour and over 15 years of hands-on market experience to every engagement.
Since 2019, we have built bespoke tools for traders and investment teams: tools they understand, own and rely on every day.
Let's talk06 / Research
A published anomaly earns 12.3% a year over sixty years of US data; seven years of Australian data give a t-statistic of 1.0. What a short sample can say, how microstructure can fake a result, and a protocol that makes either answer informative.
Read the pieceA brief literature review of attention, herding, lottery demand, the disposition effect and extrapolation, including conflicting findings.
Read the pieceInvestors who must trade leave predictable footprints. Three simulations show how the obvious measurements misread them, why competing anticipators can erase the reversal studies look for, and a one-page card that prevents both mistakes.
Read the piece“Buy” is not an instruction. A forecast curve, a liquidity profile and a cost model together determine a trading path. Three results every portfolio manager should know, and a toy optimiser that shows them.
Read the pieceThe same model can sit inside very different investment processes. A closer look at data, judgment, evaluation and implementation.
Read the pieceAn event study of intraday 8-K filings from 2015 to 2025, comparing three generations of sentiment models and what their results suggest about technology, alpha decay and trading returns.
Read the paper (PDF)Winning a forecasting test is not the same as replacing a job. What the research supports, and what it leaves unanswered.
Read the pieceAccounting identities, factor regressions and investment skill are different claims. Two synthetic examples show what reconciliation proves, how scaling changes attribution, and why a fitted residual is not evidence of alpha.
Read the pieceThe same price path can produce several different decay curves. A worked example separates marked returns, delayed entry and hindsight opportunity—and explains what a decision-relevant cost of waiting must specify.
Read the piece07 / Begin
Tell us about your process, the decisions you face and where your current tools fall short.
contact@9epoch.ai