Summer has the largest share.

Growth: six-month industrial production and three-month payroll changes from ALFRED; three-month changes in smoothed jobless claims from FRED; six-month XLI/XLP and SPY/IEF price-ratio changes from Yahoo Finance. Falling claims count as stronger growth.

Inflation: the three-month change in annual CPI inflation from ALFRED; three-month changes in five-year breakevens and six-month oil changes from FRED; six-month TIP/IEF price-ratio changes from Yahoo Finance.

Liquidity: six-month M2 growth from ALFRED. FRED supplies financial conditions (NFCI level and three-month change), the two-year Treasury yield (12 months), and Fed net liquidity (three months). Fed net liquidity subtracts reverse repos and the Treasury General Account from Fed assets, all in USD millions. Lower NFCI and falling yields count as easier conditions. L/S also includes the three-month change in high-yield spreads from FRED; narrowing spreads count as easier conditions.

From data to scores. Each reading is compared with its own trailing 120-month average and variation, including the current observation; at least 36 observations are required. Scores are capped at three standard deviations. Available signals receive equal weight within each pillar, with at least two required. The NFCI signal is half level and half change. Smoothing gives two-thirds weight to the new average and one-third to the previous smoothed score. A positive score means stronger momentum relative to that signal's history; it can occur even while an economic measure is falling.

Season sensitivity: 0.75. A smooth curve converts each Growth and Inflation score into directional shares. A zero score gives 50/50; +0.75 gives about 73/27. Multiply growth-rising by inflation-cooling for Spring, growth-rising by inflation-heating for Summer, growth-falling by inflation-heating for Fall, and growth-falling by inflation-cooling for Winter. These allocation weights total 100%. Liquidity does not change these percentages.

Each season starts with fixed ETF weights. Half the allocation follows those weights; the other half gives more weight to ETFs with lower trailing volatility. The four allocations are blended using the season percentages. Easier liquidity adds up to 15 percentage points to risk assets; tighter liquidity moves up to 25 points into defensive assets. Long-only excludes HY spreads and the credit-stress gate. HYG remains eligible as an ETF holding.

External season layer. Regime Rotation classifies month m after it closes and publishes its label on the first calendar Friday of m+1. The long-only decision at that later month-end blends a one-hot external label at 15%; when the publisher marks the label as changed, the blend is reduced to 7.5%. GSMIF uses internal Growth/Inflation seasons without this external blend, while preserving its separate 30/70 structure and 20% ETF cap. The layer never creates a first-Friday trade, uses no label before publication, and does not apply to L/S.

Grouped momentum ranks risk assets and defensive assets separately over the 12 months ending one month before the decision. Strength 0.50 gives rank multipliers from 0.75 to 1.25. Each group keeps its original allocation budget, and BIL is unchanged by momentum. At month-end, an ETF below its 150-day moving average moves to BIL in the target executed at the next NYSE-session close. An 8-month average is used when daily history is insufficient. There is no real-yield rotation between gold and Treasuries and no final portfolio-volatility scaling.

After the individual trend gates, the long-only target halves its remaining equity ETF holdings when Fall is at least 40% of the season blend or SPY closes below its 150-session average. If both are true, the reduction happens once, not twice; the released amount moves to BIL. Then, only when Spring is strictly the largest season and Liquidity is negative, 25% of post-defence US-equity exposure transfers proportionally from IWM, QQQ, SMH, SPY, USMV, VLUE, XHB, XLB, XLE, XLI, XLP, XLU and XLV into EEM. EEM cannot exceed 20% of the fund. There is no general 20% cap on non-BIL ETFs.

Both L/S macro components retain HY spreads in Liquidity. The core uses the season and liquidity blend with its own volatility limit. The enhanced component retains momentum ranked across holdings, trend and a credit-stress gate. When the HY spread exceeds 110% of its trailing 36-month median and has widened over three months, this gate halves risk-asset weights and moves that allocation to defensive holdings. Earlier history uses the Baa corporate yield less the 10-year Treasury yield. The third component takes long or short ETF positions using 12-month returns relative to BIL. Each component receives one-third of the combined allocation before the outer volatility adjustment. Separate volatility adjustments still size the components and their combination. Duplicate and opposing positions are netted before costs are charged.

Every monthly run refreshes FRED, ALFRED and Yahoo data. CPI, industrial production, payrolls and M2 use values published by each historical decision date. Other FRED series use publication delays applied to latest-revised history, so revision bias remains possible. Yahoo adjusted prices include dividends and splits. Selected early histories use mutual funds and other proxies before the ETFs existed.

Each month's signals set a target after the month-end close. The three model portfolios trade that target at the following first NYSE-session close, then it earns returns from the next session. Cut-off tests check that adding future observations leaves earlier calculations unchanged. Later research revisited both the pre-2019 and post-2019 periods, including the choice of the 150-day filter. Both are reviewed historical samples. These tests do not establish an untouched out-of-sample record.

S&P 500: CAGR 10.7%, excess Sharpe 0.50, monthly drawdown -52.9%. 60/40 SPY/AGG: CAGR 8.0%, excess Sharpe 0.57, monthly drawdown -33.2%.

Cash is the balance after ETF purchases and sales. Short collateral reserves 18.0% of portfolio value. After that reserve, the current funding deficit is 40.7%; margin interest applies to this amount. Current share counts and collateral rounding use actual dollar quotes.

Historical L/S returns include 10 basis points per unit of funded turnover, 1 basis point slippage on traded notional, margin financing and short-borrow costs. Distributions are reinvested and earlier periods use proxy histories. Current order estimates use raw quotes and per-share fees. The chart, monthly table and headline figures all use the same after-cost history.

The trend component is 146.1% gross and +43.9% net before it is combined with the other strategies. An ETF is long when its trailing 12-month return matches or beats BIL, and short when it trails BIL.

The simulation uses a USD 1 million reference account and IBKR schedules recorded on August 20, 2026. Commissions are USD 0.005 per share, with a USD 1 minimum and a 1% order-value cap, plus regulatory fees and 1 basis point of slippage. Historical interest benchmarks are estimated from Treasury-bill returns. Margin tiers start at benchmark plus 1.50%; positive free cash earns benchmark minus 0.50%, above a USD 10,000 threshold. Short-proceeds interest follows the recorded tiers, with no interest on the first USD 100,000. The model assumes 1.00% annual short-borrow cost; actual rates and availability vary.

Limits are 175% gross exposure, 25%-150% net exposure, 35% gross shorts, 25% per non-cash ETF and 50% margin debit. The account is assumed to use Portfolio Margin. IBKR must check the full portfolio before trading; the model requires a projected margin cushion of at least 20%. The 21.1% Reg T reference buffer is separate from that broker check. Direct futures roll cost is 0.0 bp; any roll effects inside ETFs are already in their returns. Current broker terms: commissions and margin.

Positions CSV · Trend positions CSV · Performance CSV · Current month CSV · Margin check CSV · Assumptions CSV