Methodology
The models, the data, and the review process behind the research, described plainly.
A slow question and a fast one
The Cycle Model
Addresses the slow question: where are we in the macro cycle? Built on a point-in-time panel of growth, inflation, liquidity, and credit data reaching back to the 1960s, it maps the current environment onto a cycle clock. Every series enters only as of its real publication date, so the historical record reflects what could actually have been known at the time.
Its role is to distinguish a pullback inside an expansion from the early stage of something larger. It sets the context for the rest of the research and anchors each Corridor issue.
The Signal Engine
Addresses the fast question: what condition is the market in right now? A weekly read across liquidity tide, trend structure, fragility, and net liquidity, normalized 0 to 100. Specific configurations of these pillars form named signals, and each signal carries a record of what followed every historical occurrence.
When the pillars disagree, the divergence itself is information. The seams between markets, and between time horizons, are where much of the useful information lives.
Model output at two historical moments
The charts below show the signal engine's stored weekly outputs, drawn from the same dataset the scorecard is computed from. Only model outputs are shown here.
Stored weekly output from the engine. Tide washed out to single digits in late March 2026 while trend held, and the late-April cross-up accompanied the recovery that followed. Tap or hover on any week for the exact readings.
The oversold-plus-liquidity configuration was active for seven straight weeks spanning the March 23, 2020 market low. Consecutive weekly occurrences cluster into episodes such as this one, which the scorecard's occurrence counts do not fully adjust for. Details are in the scorecard notes below.
Historical scorecard
Each named signal is measured on forward 13-week returns across all of its occurrences from 2008 through 2026, against an unconditional baseline of +3.1pp over 933 graded weeks. Two of the six show no reliable edge and are reported with the rest.
| Signal | Direction | Occurrences | Mean fwd 13w | Edge vs baseline | p-value |
|---|---|---|---|---|---|
| Oversold Bottom, liquidity supportive | Bullish | 42 | +8.6pp | +5.5pp | <0.001 |
| Oversold Bottom, broad | Bullish | 94 | +4.7pp | +1.6pp | 0.15 |
| Capitulation | Bullish | 4 | +4.9pp | +1.8pp | Small sample |
| Compressed Top, strict | Bearish | 30 | -0.4pp | -3.5pp | 0.01 |
| Compressed Top, broad | Bearish | 40 | -0.2pp | -3.3pp | 0.005 |
| ETF-Macro Crossdown | Bearish | 99 | +2.7pp | -0.4pp | 0.72 |
Forward returns are measured on the S&P 500 over the 13 weeks after each occurrence, 2008 through 2026. Edge is the difference from the unconditional baseline, tested with a two-sample t-test. Consecutive weekly occurrences cluster into episodes and 13-week windows overlap, which these tests do not fully adjust for. A technical note covering event clustering, overlapping horizons, multiple testing, and out-of-sample design is in preparation. Until it is published, these results should be read as historical tests rather than validated forecasts.
From question to position
1. Context. The two models establish the environment: the phase of the cycle, the condition of the market, and what has historically followed comparable conditions. Ideas are evaluated inside that context.
2. Ideas and theses. An idea can come from anywhere: an options trade, dark pool activity, an earnings report, price and volume behavior, an article, or direct experience with a product. The source of the idea is not the filter. Each idea is checked against the environment and then examined through the full research process. Ideas can start anywhere. Conviction has to survive the data.
3. Positioning. Theses become positions with sizing and risk attached, and positioning stays fluid with the market. Positions are added to, trimmed, or closed based on what the market is doing. Some conclusions will be wrong. The research process exists to make the reasoning explicit either way.
4. Deep research. Single names and specific trades are written up with a decision rule, a probability-weighted target rather than a point estimate, an account of what is already reflected in the price, and the conditions that would change the conclusion. These publish as Research Notes.
Data and models
Hanaska maintains its own research database and analytical infrastructure. The database combines historical and current information from economic, financial, derivatives, and prediction markets. Much of the underlying information is publicly available. The work is in collecting it consistently, preserving its history, connecting datasets across markets, and building models that can be tested over time.
Consistent collection
Data is collected on fixed schedules, checked for gaps and errors as it arrives, and preserved so that the historical record stays usable for research years later.
Cross-market structure
Datasets share entity and event keys, so comparing what two different markets reflected about the same moment is a query rather than a project.
Historical base rates
When a setup appears, comparable historical instances and their outcomes can be pulled from the archive. Base rates instead of anecdotes.
Machine-learning and language models are used for selected research and data-processing tasks. Models organize evidence and test hypotheses. They are inputs to the research process rather than substitutes for judgment.
How the research is published
- Research publishes on a fixed cadence and is not timed to promote any security.
- Positions held in securities discussed are disclosed in each piece.
- Published theses state the conditions that would change the conclusion.
- The public record is built from published, dated work only. Nothing is added retroactively.
- Hanaska Research publishes impersonal research and does not provide personalized investment advice.
Will Hanafan
Founder · Hanaska Research
Active investor and economics researcher, currently completing an MBA in Investment Science and an MS in Applied Economics with a concentration in Monetary and Financial Economics. Built the research system described on this page: the collectors, the database, and the models. The positions discussed in published work are real and are disclosed.