From 268ca0217a068e0bc2a81481eda4d0cf4fcb1f32 Mon Sep 17 00:00:00 2001 From: Haowei Chan Date: Wed, 21 Oct 2026 01:51:19 +0800 Subject: [PATCH] fix(propose): break a trend_add only past 3% under its ma50 One close under the MA50 exited a trend_add on noise: on 181 sampled US names 2014-2026 that rule exited 9.1x a year and 59% of exits closed back above within 5 sessions. 3% under cuts it to 4.0x and 25%, median CAGR 3.7% -> 5.3%, same median drawdown. signals.MA_TREND_BUFFER holds the number. Co-Authored-By: Claude Opus 5.5 --- README.md | 4 ++-- docs/ARCHITECTURE.md | 3 ++- kuroshio/core/allocator/engine.py | 20 +++++++++++--------- kuroshio/core/allocator/signals.py | 5 +++++ tests/test_allocator.py | 14 +++++++++++++- 5 files changed, 33 insertions(+), 13 deletions(-) diff --git a/README.md b/README.md index db0ac38..2d102d9 100644 --- a/README.md +++ b/README.md @@ -72,8 +72,8 @@ different scales. Record *why* you own a position — `setup_type:` (value_dip | pullback_add | trend_add | other) plus `entry_price:` and, for the dip setups, `invalidation_price:` — and `propose` -watches each one on its own terms: a `trend_add` is alerted when it closes under its 50-day -mean, a `value_dip` never is (looking weak against its MAs is the setup) and is alerted only +watches each one on its own terms: a `trend_add` is alerted when it closes more than 3% under +its 50-day mean, a `value_dip` never is (looking weak against its MAs is the setup) and is alerted only when it closes at or below the invalidation price you recorded. Add an `entry_date:` and the stop moves with the tape: a `trend_add` — and a `pullback_add` once its running high has cleared entry + 2R — trails its invalidation price at `caps.trail_atr_mult` ATR14s below the diff --git a/docs/ARCHITECTURE.md b/docs/ARCHITECTURE.md index 375f56a..f0c83dc 100644 --- a/docs/ARCHITECTURE.md +++ b/docs/ARCHITECTURE.md @@ -198,7 +198,8 @@ Pure function. v1 logic: letting cash absorb the difference. 3. **Thesis monitoring**, dispatched on `setup_type` — the ranking in step 4 is a momentum composite, so a `value_dip` is *meant* to look weak there and must not be alerted on that: - `trend_add` → ALERT when the close is under its 50-day mean (the trend was the thesis); + `trend_add` → ALERT when the close is more than 3% under its 50-day mean (the trend was the + thesis; `signals.MA_TREND_BUFFER` keeps one close under the line from being a break); `value_dip` / `pullback_add` → ALERT only when the close is at or below the `invalidation_price` the user recorded, never on MA distance. Cards name the setup_type, the level breached, the entry price with the move from it, and the session the price came diff --git a/kuroshio/core/allocator/engine.py b/kuroshio/core/allocator/engine.py index 36654a4..2f77530 100644 --- a/kuroshio/core/allocator/engine.py +++ b/kuroshio/core/allocator/engine.py @@ -16,7 +16,7 @@ from decimal import Decimal -from kuroshio.core.allocator.signals import BOOK_VOL_WINDOW, MA_TREND +from kuroshio.core.allocator.signals import BOOK_VOL_WINDOW, MA_TREND, MA_TREND_BUFFER from kuroshio.types import Candidate, Holding, ProposalCard # setup_types that carry a monitoring rule. "other" (and a missing setup_type) carry @@ -85,12 +85,12 @@ "entry_missing": "entry price not recorded", "no_ma50": "no MA50 for its trend_add — fewer than {ma_trend} traded sessions", "trend_intact": ( - "its trend is intact — {at}, at or above its 50-day moving " - "average of {ma:.2f}" + "its trend is intact — {at}, not more than {buffer:.0%} under its 50-day " + "moving average of {ma:.2f}" ), "trend_broken": ( "{ticker} was opened as a trend_add and the trend has broken: {at}, " - "below its 50-day moving average of {ma:.2f} ({entry}). " + "more than {buffer:.0%} under its 50-day moving average of {ma:.2f} ({entry}). " "The setup that justified the position no longer holds." ), "trail_known": "{stop:.2f} its stop has ratcheted up to (see the ALERT above)", @@ -251,10 +251,10 @@ "entry_recorded": "進場價 {entry_price:.2f}，目前距進場 {chg:+.1%}", "entry_missing": "沒有記錄進場價", "no_ma50": "這檔趨勢加碼沒有 MA50 — 交易日數不到 {ma_trend} 天", - "trend_intact": "趨勢仍然成立 — {at}，現價在 50 日均線 {ma:.2f} 之上或持平", + "trend_intact": "趨勢仍然成立 — {at}，現價沒有低於 50 日均線 {ma:.2f} 超過 {buffer:.0%}", "trend_broken": ( "{ticker} 當初以趨勢加碼進場，趨勢已經走壞：{at}，低於 50 日均線 " - "{ma:.2f}（{entry}）。當初進場的理由已經不成立。" + "{ma:.2f} 超過 {buffer:.0%}（{entry}）。當初進場的理由已經不成立。" ), "trail_known": "{stop:.2f}（這次檢查剛把停損上調到這裡，見上方警示）", "trail_earlier": "{stop:.2f}（之前的檢查已經把停損上調到這裡）", @@ -751,10 +751,12 @@ def cap_for(theme: str) -> tuple[float, str]: if ma is None: thesis_gap[h.ticker] = T["no_ma50"].format(ma_trend=MA_TREND) continue - if price >= ma: - thesis_note[h.ticker] = T["trend_intact"].format(at=at, ma=ma) + if price >= ma * (1 - MA_TREND_BUFFER): + thesis_note[h.ticker] = T["trend_intact"].format(at=at, ma=ma, buffer=MA_TREND_BUFFER) continue - reason = T["trend_broken"].format(ticker=h.ticker, at=at, ma=ma, entry=entry) + reason = T["trend_broken"].format( + ticker=h.ticker, at=at, ma=ma, buffer=MA_TREND_BUFFER, entry=entry, + ) details = {"ma50": ma} elif h.setup_type == "trend_add": reason = T["trail_breached"].format(ticker=h.ticker, at=at, level=level, entry=entry) diff --git a/kuroshio/core/allocator/signals.py b/kuroshio/core/allocator/signals.py index f5cbe42..43c7fa5 100644 --- a/kuroshio/core/allocator/signals.py +++ b/kuroshio/core/allocator/signals.py @@ -18,6 +18,11 @@ from kuroshio.types import Holding, Panel MA_TREND = 50 # the trend_add break window; see engine.propose step 3 +# A trend_add breaks only on a close more than this far under its MA50, not on any close under +# it. On 181 sampled US names 2014-2026, one close under the line exited 9.1x a year with 59% +# back above within 5 sessions; 3% under cut that to 4.0x and 25%, median CAGR 3.7% -> 5.3%, +# same median drawdown. Deeper keeps trading exits for a looser rule (5%: 2.7x, 14%). +MA_TREND_BUFFER = 0.03 BOOK_VOL_WINDOW = 20 # trailing-session window book_vol() and engine.propose step 2b share ATR_WINDOW = 14 # the ATR the stop ratchet trails by; see engine.propose step 3 diff --git a/tests/test_allocator.py b/tests/test_allocator.py index 113e2e6..e80c258 100644 --- a/tests/test_allocator.py +++ b/tests/test_allocator.py @@ -296,6 +296,18 @@ def test_trend_add_alerts_on_ma_break_and_the_dip_setups_do_not(): assert set(thesis_alerts(cards)) == {"TREND"} +def test_trend_add_breaks_only_past_the_ma50_buffer(): + """One close a little under the MA50 is noise, not a break: 97.00 against 100.00 holds, + 96.99 is more than MA_TREND_BUFFER under and EXITs.""" + def exited(price): + cards = propose(thesis_portfolio(), [], make_ips(), "us", + prices={"TREND": price}, ma50={"TREND": 100.0}) + return "TREND" in thesis_alerts(cards) + assert not exited(99.0) + assert not exited(97.0) + assert exited(96.99) + + def test_value_dip_alerts_on_invalidation_breach(): """Acceptance clause 3: the value_dip's own invalidation price is what ends it — and the pullback_add answers to the same rule.""" @@ -489,7 +501,7 @@ def test_swap_selling_an_intact_trend_add_says_the_trend_is_intact(): swap = next(c for c in cards if c.action == "SWAP") assert swap.sell == "TREND" assert "its trend is intact" in swap.reason - assert "at or above its 50-day moving average of 100.00" in swap.reason + assert "not more than 3% under its 50-day moving average of 100.00" in swap.reason assert "broken" not in swap.reason