How Limnal builds an allocation
Limnal reads public economic and market data daily, scores the macro regime, and translates the score into a portfolio allocation. The pipeline is deterministic and rules-based.
What goes into the assessment
Three families of public-domain data feed the models daily. Macro: Treasury yields and yield-curve shape, TIPS-derived inflation expectations, and high-yield credit spreads (FRED), plus the US Treasury Department's Daily Treasury Statement cash flows. Valuation: Robert Shiller's US Cyclically-Adjusted P/E (CAPE) and Research Affiliates' non-US CAPE across developed and emerging markets. Price action: sector ETF price history and momentum across equity, fixed income, commodities, and alternatives (Yahoo Finance), realized and implied volatility, and futures-market positioning from the CFTC's Commitments of Traders. None of the inputs are proprietary. The composition is.
Multi-strategy consensus
Six quantitative strategies score the regime daily on a 1.0 (risk-off) → 3.0 (risk-on) scale. Their consensus drives the allocation. Each strategy reads a distinct data stream; agreement and disagreement are both signal.
Systematic, not AI
Every input feeds a deterministic calculation. The same inputs produce the same allocation. Limnal does not use machine learning or generative AI anywhere in the model.
Per-asset-class scoring
Each sleeve (equity, fixed income, commodities, alts) gets a score by averaging the strategies most relevant to it. The score sets the sleeve's share of the portfolio. Cash takes the residual when scores are weak; strong scores pull cash back into risk assets.
Five rebalance cadences
Weekly rebalances every Monday. Monthly, Quarterly, Semi-Annual, and Annual rebalance on their respective calendar boundaries (first trading day of each month, quarter, half-year, or year), plus an off-cycle rebalance whenever a sleeve drifts meaningfully from its current weight. Slower cadences trade less and incur less tax drag; faster cadences track regime shifts more responsively. A daily kill switch runs underneath all five cadences and sharply reduces equity exposure when trend and momentum signals both confirm bearish. Adaptive+variants add a leveraged-equity overlay when risk-on signals align; the overlay's gating is independent of the cadence.
Match cadence to account type
The backtest is frictionless: no taxes on realized gains. Real money pays them. Weekly and the leveraged Adaptive+ variants belong in a tax-advantaged account (Roth IRA, traditional IRA, 401(k) rollover) — frequent adjustments otherwise trigger short-term gains. Quarterly, Semi-Annual, and Annual work in a regular taxable account: most realized gains end up long-term. Monthly sits in between. The welcome page has the full cadence-by-account-type matrix.
Point-in-time inputs (no look-ahead)
Every historical model run uses only data that was available on that as-of date. Yields, valuations, and prices are queried with the cutoff in place; future data is invisible to the simulation. This rules out look-ahead bias — the most common way backtests overstate performance.
Benchmarks
100% SPY (passive equity); 60/40 SPY/AGG (rebalanced weekly); VFIFX (Vanguard Target Retirement 2050, buy-and-hold of the fund's NAV; the fund's internal glide path is already baked in). All three are real, investable portfolios — not abstract indices.
Frictionless backtest
No transaction costs, bid-ask spreads, or slippage. A real weekly-rebalance implementation across a diversified ETF book gives back roughly 30–50 basis points per year to friction; longer cadences give back less. Read the published numbers as an upper bound on realized performance, not a forecast.
Window
The published backtest runs from January 2016 to today — roughly 10 years of daily data. We start in 2016 because that is the period over which the full model suite has genuine point-in-time inputs; the Treasury-flow and dealer-flow models in particular have no reliable history before then, and extending earlier would mean scoring models on data they never actually saw. The window covers the late-2010s bull market, the 2018 vol-mageddon, the 2020 COVID drawdown and recovery, the 2022 bear market, and the post-2022 AI rally. It extends as live data accumulates.
Trend signal families
The Momentum & Trend model combines three signals on each ETF: direction (which way price is moving), volatility (how turbulent the move is), and strength (how clean the trend is relative to its own recent history). Direction and strength confirm an uptrend; volatility sets how much weight to put on it. Per-ETF outputs roll up into sector and asset-class scores.
This is information, not investment advice. Backtests describe what the model would have done with the data available at each historical point in time. They do not guarantee future results. The published numbers reflect a frictionless simulation; real transaction costs, taxes, and slippage will subtract from realized returns. Limnal Research provides analytics. Investment decisions are yours.
See the performance page for the backtest results across cadences, or the welcome page for the broader product overview.