# Taxable KISS stock-overlay research

## Decision status

Research only. No Taxable KISS production target, feed, methodology version, or public performance claim changes as a result of this document.

The current `taxable-kiss-5.2.0` stock overlay does not clear its stated secular-drawdown objective in the 1960s--1980s price-history test. It missed the 1973--74 drawdown because the July 1972 defensive state reversed before the decline. The threshold requiring an exact bearish cloud cross while price remains materially above its 200-week average makes the overlay a late-cycle trim rather than a persistent secular-bear defense.

## Research question and frozen candidates

The research starts with a 60% defensive stock weight because that is already the published Taxable KISS defensive exposure. The following candidates were frozen before running the monthly total-return comparison:

1. **Monthly trend state:** after two consecutive monthly S&P 500 observations below their 10-month average, reduce stocks to 60%; after two consecutive observations above the average, restore 100%.
2. **Stricter recovery:** use the same reduction rule but do not restore normal exposure until the trough observed during the defensive state is at least six months old as well as the two-close recovery condition being met.
3. **Annual valuation cap:** apply the monthly trend state, but after a December CAPE observation set the following calendar year's stock maximum to 60% when CAPE is at or above 20, 25, or 30. A defensive state is then 60% of that maximum (36% stocks).
4. **Low-frequency bear protocol:** after two consecutive monthly observations below the 10-month average *and* a 10% or 15% decline from the trailing 12-month high, reduce stocks to 20%. Restore 100% after two consecutive observations above the 10-month average. This is deliberately a one-way state change, not a series of incremental trades.

Each rule is evaluated at a month-end observation and implemented for the following month. A 10-basis-point cost is charged whenever the stock target changes. This is still a **pretax** simulation: no lot, realized-gain, wash-sale, distribution, or state-tax treatment is modeled.

## Data and return construction

- The monthly equity data are the public [Shiller S&P 500 dataset](https://github.com/datasets/s-and-p-500), which reports price, trailing annual dividend, CPI, and CAPE. Its historical price series is monthly average daily closes, not a month-end ETF total-return index.
- The stock return approximation is monthly price change plus one-twelfth of the reported trailing annual dividend.
- Cash earns the preceding month's [FRED TB3MS](https://fred.stlouisfed.org/series/TB3MS) annualized yield divided by twelve.
- The comparison is intentionally a stock-sleeve study. It does not reconstruct a pre-1975 gold implementation or a pre-bitcoin full Taxable KISS portfolio.

This construction is appropriate for ranking simple, slow rules but is insufficient for a production claim. A promotion candidate needs investable, adjusted-total-return data, point-in-time data validation, an explicit taxable-lot simulation, parameter-neighborhood analysis, and paper trading.

## Results

All returns below are nominal total-return approximations, net of the declared 10-basis-point target-change cost. Maximum drawdown is based on monthly observations, so it is less severe than a daily-close measurement.

| Window | Monthly trend 60% | Buy and hold | Trend result |
|---|---:|---:|---|
| Feb. 1966--Aug. 1982 return | +179.3% | +129.5% | Higher return |
| Feb. 1966--Aug. 1982 maximum drawdown | -24.8% | -39.2% | 14.4 points less drawdown |
| Jan. 1973--Oct. 1974 return | -23.5% | -37.5% | Meaningful protection |
| Jan. 1973--Oct. 1974 maximum drawdown | -24.8% | -39.0% | Meaningful protection |
| 1983--2018 return | +4,219.2% | +4,104.3% | Similar return |
| 1983--2018 maximum drawdown | -33.1% | -49.0% | 15.9 points less drawdown |
| 2000--2009 return | +31.0% | -6.9% | Better in two major bears |
| 2010--2018 return | +140.7% | +172.9% | Underinvested in a sustained bull |
| 1960--2018 return | +27,081.8% | +24,948.5% | Higher return |
| 1960--2018 maximum drawdown | -33.1% | -49.0% | Lower drawdown |

The monthly-trend candidate made 24 reductions in the 59-year 1960--2018 sample: 0.41 reductions per year. It is therefore modestly more active than the current exact-cross overlay, but remains below the one-sale-per-asset-per-year research objective.

### Low-frequency left-tail bear protocol: advance

The gentler 60% state cannot protect against the stated risk: after a 10--20% trigger, it still leaves 60% of the stock sleeve exposed to a potential 70--80% collapse. The relevant test is the 20% stock state, triggered only after both a persistent monthly trend break and a 10% decline from the trailing 12-month high.

| Window | 20% bear protocol | Buy and hold | Bear-protocol result |
|---|---:|---:|---|
| Feb. 1966--Aug. 1982 return | +210.9% | +129.5% | Higher return |
| Feb. 1966--Aug. 1982 maximum drawdown | -11.9% | -39.2% | 27.3 points less drawdown |
| Jan. 1973--Oct. 1974 return | -9.0% | -37.5% | Material protection |
| Jan. 1973--Oct. 1974 maximum drawdown | -10.4% | -39.0% | 28.6 points less drawdown |
| 1983--2018 return | +4,947.5% | +4,104.3% | Higher return |
| 1983--2018 maximum drawdown | -25.0% | -49.0% | 24.0 points less drawdown |
| 2000--2009 return | +69.4% | -6.9% | Better in two major bears |
| 2000--2009 maximum drawdown | -18.3% | -49.0% | Material protection |
| 2010--2018 return | +141.7% | +172.9% | The expected sustained-bull opportunity cost |
| 2019--June 2023 untouched return | +53.2% | +79.4% | Trailed buy and hold |
| 2019--June 2023 maximum drawdown | -18.9% | -19.3% | No material protection |
| 1960--2018 return | +34,020.7% | +24,948.5% | Higher return |
| 1960--2018 maximum drawdown | -25.0% | -49.0% | Lower drawdown |

It reduced stocks in July 1973, after an 11.5% trailing-high decline, and restored them in April 1975 after the two-month trend recovery. It made 13 reductions in 59 years: 0.22 per year. It is therefore less active than the current exact-cross stock rule while making a large, economically meaningful reduction when the trigger does occur.

This rule cannot protect a one-month crash such as October 1987: a month-end system can act only after the loss has occurred. The previously untouched 2019--June 2023 window illustrates that limitation: it provided no material monthly drawdown protection through the COVID shock and trailed buy and hold. It is a low-frequency secular-bear protocol, not crash insurance. A 15% threshold reduced turnover to eight reductions but produced materially weaker 1970s protection; the 10% threshold is the candidate to take forward for further research, not for promotion.

### Stricter recovery: reject

The six-month trough-age requirement did **not** improve maximum drawdown in any of the tested secular-bear windows. With the 60% trend state it remained -24.8% in the 1966--82 and 1973--74 tests, while 1960--2018 terminal return fell from +27,081.8% to +18,410.9%. With the 20% bear protocol, the stricter recovery produced the same 1960--2018 maximum drawdown (-25.0%) as the ordinary two-close recovery but cut terminal return from +34,020.7% to +15,072.2%. It makes the model miss recoveries without delivering extra left-tail protection. Reject it.

### Locked Bitcoin extrapolation: reject as a shared bubble rule

As an out-of-sample check, the 10-month/two-close/10%-trailing-high/20% rule was applied to Bitcoin **without changing a parameter**. This does not change the existing Bitcoin Taxable KISS Ichimoku/200-week-average policy and is not a realistic bitcoin total-return simulation. It uses the public [Blockchain.com market-price series](https://www.blockchain.com/explorer/charts/market-price), converted to the last available observation in each month, with the same T-bill cash convention.

| Bitcoin window | Locked protocol maximum drawdown | Buy-and-hold maximum drawdown | Trigger timing |
|---|---:|---:|---|
| Dec. 2013--Jan. 2015 | -42.7% | -66.5% | Entered only after a 55.3% trailing-high decline |
| Dec. 2017--Dec. 2018 | -55.7% | -70.1% | Entered only after a 53.4% trailing-high decline |
| Nov. 2021--Nov. 2022 | -38.9% | -70.0% | Entered only after a 38.6% trailing-high decline |

The rule did reduce the remaining portion of each decline, but it plainly did not exit after a 10--20% bitcoin decline. Bitcoin's normal monthly volatility and sharp initial reversals make a two-month slow-trend confirmation too late for that asset. This is useful negative evidence: do not tune this stock protocol to Bitcoin after seeing these results, and do not replace the existing Bitcoin-specific rule with it.

### Valuation cap: do not advance

The CAPE cap is not persuasive as an addition to the trend rule:

| Candidate | 1966--82 max drawdown | 1983--2018 max drawdown | 1960--2018 return | Main issue |
|---|---:|---:|---:|---|
| Trend only | -24.8% | -33.1% | +27,081.8% | Control |
| CAPE >= 20 | -24.8% | -25.5% | +17,588.3% | Large long-run opportunity cost |
| CAPE >= 25 | -24.8% | -25.5% | +22,632.4% | No added 1960s--70s protection |
| CAPE >= 30 | -24.8% | -33.1% | +28,493.3% | No additional drawdown protection |

The 20 threshold helps later drawdowns only by permanently lowering the stock maximum for extended, expensive periods; it sharply trails in the post-2010 bull. The 25 and 30 thresholds fail to improve the 1960s--70s maximum drawdown beyond the simple trend rule. More importantly, the public CAPE history is not point-in-time vintage data and a threshold selected after inspecting this history would be vulnerable to overfitting. The evidence does not support adding a valuation cap to Taxable KISS.

### 25% stocks / 25% bitcoin / 25% gold / 25% cash: annual rebalance remains the control

The stock protocol was then applied to a hypothetical equal-weight portfolio. Bitcoin and gold were left unchanged; only the stock sleeve could fall from 25% to 5%, with the released 20% retained as cash. The comparison rebalances the ordinary portfolio annually and rebalances the protocol portfolio annually subject to its active stock target. Gold and Bitcoin use monthly price proxies, so this is a directional portfolio test rather than a production performance claim.

| Window | Annual 25/25/25/25 | Stock bear protocol | Interpretation |
|---|---:|---:|---|
| 2011--June 2023 return | +98,558.8% | +95,447.5% | Annual rebalance higher |
| 2011--June 2023 maximum drawdown | -78.1% | -78.2% | No portfolio-level protection |
| 2013 bitcoin-bubble window maximum drawdown | -15.8% | -15.8% | No stock protocol activation |
| 2017 bitcoin-bubble window maximum drawdown | -18.0% | -18.0% | No stock protocol activation |
| 2021--22 window maximum drawdown | -23.8% | -23.3% | Only a modest benefit |

The stock protocol made three reductions in this sample. It did not reduce the portfolio's maximum drawdown because the dominant loss came from Bitcoin, not equities. Annual rebalancing captured more of Bitcoin's subsequent recoveries. The correct conclusion is not that diversification always eliminates equity-tail risk; it is that a stock-only bear overlay has to earn its place on **portfolio** expected shortfall and drawdown, not merely on a standalone equity-sleeve test.

For an equal-weight portfolio with an asset-specific Bitcoin protocol already in force, annual rebalancing should remain the benchmark and default. Add a stock bear protocol only if it improves the full portfolio after Bitcoin's existing protection, trading costs, and taxable implementation are modeled. The current evidence does not establish that benefit.

### One-third stocks / one-third gold / one-third cash: overlay is a modest drawdown improvement

Removing Bitcoin isolates the stock question. Both variants below rebalance annually. The second also applies the locked stock bear protocol, reducing stocks from one-third to one-fifteenth of the portfolio (20% of their strategic maximum) and holding the released amount in cash.

| Window | Annual rebalance return | Stock-overlay return | Annual maximum drawdown | Stock-overlay maximum drawdown |
|---|---:|---:|---:|---:|
| Feb. 1966--Aug. 1982 | +438.1% | +465.8% | -16.6% | -14.4% |
| 1980--1989 | +106.1% | +92.2% | -15.4% | -13.1% |
| 2000--2009 | +81.3% | +113.7% | -17.2% | -9.7% |
| 2019--June 2023 untouched | +45.1% | +38.5% | -9.4% | -7.6% |
| 1960--2018 | +8,006.1% | +8,029.1% | -17.2% | -14.4% |

The overlay made 12 stock reductions between 1960 and 2018 (0.20 per year). It lowered drawdown in every listed window, but its return advantage comes from the 1960s--70s and 2000s bear periods; it trailed during the 1980s and the post-2018 holdout. The diversified portfolio already absorbs much of the equity left tail through gold and cash, so the overlay is a modest incremental risk control, not a dominant return engine.

For a taxable account, that modest pre-tax benefit is not sufficient for adoption by itself. It must survive an after-tax lot simulation and a daily investable-instrument replication. Pre-1971 gold pricing also includes the Bretton Woods fixed-price period, and physical gold was not freely investable for U.S. investors until 1975; treat the 1960s--70s portfolio result as historical context, not an executable U.S. portfolio record.

### 60% stocks / 30% gold / 10% cash: annual rebalance plus the stock protocol is the leading proxy

This comparison removes Bitcoin and starts all portfolios at 60% stocks, 30% gold, and 10% cash. It compares:

1. annual rebalancing only;
2. the **existing Taxable KISS stock Ichimoku overlay** with Bitcoin removed and no annual sleeve rebalance (a research proxy, not published Taxable KISS performance);
3. the proposed 20% stock bear protocol without annual sleeve rebalance; and
4. annual rebalance plus the proposed stock bear protocol.

| 1960--2018 proxy result | Annual only | Existing stock overlay | New overlay, no annual rebalance | Annual + new stock protocol |
|---|---:|---:|---:|---:|
| Total return | +17,006.6% | +9,373.2% | +14,458.4% | **+18,052.7%** |
| CAGR | 9.12% | 8.03% | 8.82% | **9.23%** |
| Sortino | 1.77 | 1.52 | 1.88 | **2.30** |
| Maximum drawdown | -26.4% | -36.7% | -31.4% | **-17.7%** |
| Stock reductions | 0 | 18 | 13 | 13 |

The proposed overlay does **not** justify abandoning annual rebalancing. The annual rebalance is what realizes the diversification benefit of the gold and cash sleeves; without it, both overlays have worse maximum drawdown than the annual-only portfolio. With it, the new protocol is the leading result in the long proxy sample.

| Window | Annual only: total / CAGR / Sortino / max drawdown | Annual + new protocol: total / CAGR / Sortino / max drawdown |
|---|---|---|
| Feb. 1966--Aug. 1982 | +378.4% / 9.95% / 1.53 / -22.0% | **+435.8% / 10.71% / 2.06 / -17.7%** |
| 1980--1989 | **+159.0% / 10.07% / 1.61 / -21.2%** | +128.7% / 8.70% / 1.50 / -16.8% |
| 2000--2009 | +59.1% / 4.80% / 0.72 / -26.4% | **+119.2% / 8.24% / 2.35 / -10.7%** |
| 2019--June 2023 untouched | **+64.2% / 11.89% / 1.71 / -13.8%** | +56.7% / 10.70% / 1.68 / -11.3% |

The trade-off is now explicit: the protocol meaningfully improves secular-bear and 2000s outcomes but trails in the 1980s and the post-2018 holdout. The stock sleeve should not be permanently invested merely because gold is strategic. Rather, the evidence supports a **rare, substantial, stock-only defensive state paired with annual strategic rebalancing**, subject to the remaining tax and daily-data validation.

## Interpretation

Slow trend works here because it stays defensive during the persistent decline: it reduced after May 1973 and did not restore full exposure until April 1975. The current Ichimoku stock rule instead restored full exposure in August 1972, before the major drawdown began. The slow rule does not protect against a one-month crash; it is a secular-trend overlay, not crash insurance.

Long-horizon time-series trend has broad historical support as a category, including across many crisis windows in [Hurst, Ooi, and Pedersen (2017)](https://research.cbs.dk/en/publications/a-century-of-evidence-on-trend-following-investing/). That evidence supports a careful test, not a claim that these exact parameters will work going forward.

## Recommendation

Advance only the monthly 10-month/two-close/10%-trailing-high/20%-stock bear protocol to the next research stage. Keep its rule set exactly as described above while testing:

1. investable total-return equity data with daily next-session execution;
2. a prespecified 1950--1989 development interval and 1990--2018 holdout, followed by a post-2018 untouched check;
3. adjacent 8/10/12-month averages, 8%/10%/12% drawdown triggers, one/two/three-close confirmations, and 0%/20%/40% defensive weights;
4. explicit taxable-lot outcomes under several tax rates; and
5. paper-traded signals before any methodology promotion.

Reject the six-month stricter re-entry rule and the annual CAPE cap from the active candidate set. If the bear protocol survives the next stage and is promoted, retire the stock Ichimoku overlay rather than retaining it as a diagnostic allocation input.

## Reproduction

Download the public source CSVs, then run:

```sh
node scripts/analyze-taxable-kiss-stock-history.mjs \
  --shiller=/path/to/shiller-sp500.csv \
  --tbill=/path/to/tb3ms.csv \
  --bitcoin=/path/to/bitcoin-market-price.json \
  --gold=/path/to/gold-monthly.csv \
  --daily-stock=/path/to/sp500-daily.csv
```

The optional Bitcoin input runs the locked out-of-sample extrapolation. Supplying gold and daily-stock inputs also runs the 60/30/10 stocks/gold/cash comparison, including the current stock Ichimoku proxy, annual rebalancing, and the proposed protocol. The script emits the full decision ledger and all frozen-window metrics as JSON. It does not download data, write generated artifacts, or change production portfolio data.