入门Start here组合Portfolio
Asset allocation is not one answer: from 60/40 to All Weather and L1资产配置不是一道选择题：从 60/40 到全天候，再到 L1
2026-11-05 · 8 min
把现有资产配置研究整理成一条决策路径：先用 60/40 等基准定义比较尺子，再用永久组合、全天候、风险平价和 1/N 讨论风险分配，随后用再平衡、约束和负债口径处理执行，最后以趋势覆盖与 L1 影子组合形成可归因、可复核的闭环。专题已覆盖主要策略族，下一步不是继续堆文章，而是统一比较口径并保留持续前向记录。
A decision map through the existing allocation research: benchmark selection, risk allocation, implementation constraints and an auditable process. It connects the 60/40, Permanent Portfolio, All Weather, risk parity, 1/N, rebalancing, constraints, liability-driven allocation, trend following and the L1 shadow portfolio.
组合Portfolio归因Attribution数据与工程Data & Engineering
From Views to Weights: the benchmark won. What survives is the process.从观点到组合：基准赢了，流程留下了什么？
2026-10-30 · 14 min
一个公开多资产影子组合的完整研究案例：8 个 ETF、221 个月、月度再平衡和 10 bps 换手成本。L1 TAA 年化 5.20%，同期 60/40 为 8.43%——基准赢了；但冻结政策、风险贡献、收益归因和负结果留痕，构成一套可复核的投资流程。本次同步修复收益贡献占位符，并加入五分钟英文投资陈述入口。
A public multi-asset shadow portfolio: eight ETFs, 221 months, monthly rebalancing and 10 bps turnover costs. L1 TAA earned 5.20% a year against 8.43% for a 60/40 — the benchmark won. The durable output is the auditable process, now paired with a five-minute English presentation.
组合Portfolio
Rebalancing, tested: how many basis points a year, and how wide a band再平衡实测：每年值多少个基点，容差带该设多宽
2026-10-12 · 19 min
「再平衡能提高长期收益」一半对一半错。用 1993-02 至 2025-09 共 392 个月的美国数据、8 条腿、4 个目标组合（2/4/8/8 条腿）与 8 种再平衡政策：分散化收益与 Booth–Fama 公式几乎完全吻合（60/40 理论 0.35% 对实测 0.33%，1/N 八腿 0.85% 对 0.88%），且随腿数上升；但同一份数据里不再平衡比月频再平衡的年化高 0.45–1.17 个百分点——60/40 三十年后漂到 91.9% 是股票。按夏普排，年度再平衡在四个组合里全部胜出。成本从 1 倍放大到 10 倍只再多损失 4–42 个基点；容差带从 1% 拉到 28% 换手单调下降，溢价却没有一致最优宽度。含 9 张图表、1 张可拖带宽的交互图与 4 张表。
Half of "rebalancing improves long-run returns" is right and half is wrong. On 392 monthly US observations from 1993-02 to 2025-09, across eight legs, four target portfolios (two, four, eight and eight legs) and eight policies: the diversification return matches the Booth–Fama formula almost exactly (0.35% predicted against 0.33% measured for a 60/40; 0.85% against 0.88% for 1/N) and rises with the number of legs. But over the same data buy-and-hold beats monthly rebalancing by 0.45 to 1.17 points a year — a 60/40 drifts to 91.9% equities. Ranked by Sharpe, annual rebalancing wins in all four portfolios. Scaling costs tenfold costs only another 4–42 basis points, and widening a tolerance band from 1% to 28% lowers turnover monotonically without producing a consistent optimum.
组合Portfolio
Liability-driven allocation: change the denominator and the ranking changes负债驱动的资产配置：换一个分母，排序就变了
2026-10-09 · 20 min
把一笔 20 年期定额年金（按当月 10 年期美国国债收益率折现、修正久期均值 9.00 年）当作负债，用 1993-02 至 2025-09 共 392 个月的美国数据检验七个组合。资产端夏普第一的 1/N 等权（0.71）在盈余端排第六；资产端垫底的久期匹配（0.26）盈余波动只有 2.15%、funding ratio 最大回撤只有 −2.4%。久期缺口与盈余波动是一条 U 形曲线，最低点在资产久期等于负债久期处（7.54% → 2.19%）。利率动 2 个百分点，缺口 −9 年的纯股组合 funding ratio 动 18%。可承受的风险随 funding ratio 系统性放宽：100% 时连 5% 的跌破容忍度都只能持有 0% 股票，115% 时可以持有 55%。含 9 张图表、1 张可拖股票权重与久期目标的交互图与 4 张表。
A 20-year level annuity (discounted at the 10-year US Treasury yield, average modified duration 9.00 years) becomes the liability for seven portfolios over 392 monthly US observations from February 1993 to September 2025. The best asset-only portfolio, 1/N at a 0.71 Sharpe, ranks sixth on surplus; the worst, duration matching at 0.26, delivers 2.15% surplus volatility and a −2.4% maximum funding-ratio drawdown against −58.9% for all-equities. The duration gap and surplus volatility trace a U whose minimum sits where asset duration equals liability duration (7.54% to 2.19%), and a two-point rate move shifts an all-equity portfolio's funding ratio by 18%. The tolerable risk budget widens systematically with the funding ratio: at 100% even a 5% shortfall tolerance permits zero equities, while 115% permits 55%.
组合Portfolio
1/N beats the optimiser: eight legs, nine portfolios, thirty-three years1/N 打败最优化：八条腿、九个组合、三十三年
2026-10-08 · 19 min
用 1990-02 至 2025-09 共 428 个月（样本外 392 个月）的美国数据，把 1/N 等权与八个组合优化器放在同一套规则下重跑。1/N 年化 5.88%、波动 6.4%、最大回撤 −25.1%、夏普 0.54；八个优化器里三个赢（风险平价 ERC 0.65、最大夏普·收缩 0.58、逆波动率 0.57）、一个打平、四个输。样本内预期夏普平均 1.74 的最大夏普组合，实际拿到 0.54——与什么都不算的等权完全相同，九个方法里预期与实现的相关性最好只有 0.04。收缩估计能把最小方差从 0.11 修到 0.38，修不到 1/N 之上；换估计窗口（12–120 个月）与换资产数（2–8 条腿）都不改变排序。优化器的价值是条件性的：2008 年最小方差只跌 0.7%（1/N 跌 24.6%），高通胀三年里八个优化器只有一个赚钱。含 11 张图表（其中 1 张可拖估计窗口）与 4 张表。
428 monthly US observations from February 1990 to September 2025 (392 out of sample) re-run 1/N against eight portfolio optimisers under one set of rules. 1/N earns 5.88% a year with 6.4% volatility, a −25.1% maximum drawdown and a 0.54 Sharpe; three optimisers beat it (risk-parity ERC 0.65, shrunk max-Sharpe 0.58, inverse volatility 0.57), one ties and four lose. The portfolio whose in-sample expected Sharpe averaged 1.74 delivered 0.54 — identical to equal weighting, which estimates nothing — and across nine methods the best promise-to-delivery correlation is 0.04. Shrinkage lifts minimum variance from 0.11 to 0.38 but not past 1/N, and neither the estimation window (12–120 months) nor the number of legs (2–8) changes the ordering. The optimiser's value is conditional: minimum variance lost 0.7% in 2008 against 1/N's 24.6%, and in three years of high inflation only one of the eight made money.
组合Portfolio
Risk parity, tested: the risk gets flattened, the return does not follow风险平价实测：风险被拍平了，收益没有跟上来
2026-10-08 · 18 min
不按资金分配、按风险分配——这条原则经得起检验吗？用 1993-02 至 2025-09 共 392 个月的美国数据，把股票、长债、黄金、商品四条腿做成等风险贡献组合（每年 1 月按最近 36 个月协方差重解权重）：32.7 年年化 7.40%、波动 7.2%、最大回撤 −22.5%、夏普 0.69。机制确实成立——60/40 里 60% 的钱拿走 93.5% 的风险，风险平价把它拍到 27.3 / 21.7 / 24.5 / 26.6；但收益没有跟上：它比 60/40 少 1.3 个百分点年化，甚至输给最笨的 25/25/25/25（夏普 0.73 对 0.69）。要补回这 1.3 个点就得上杠杆，而 2022 年杠杆版跌得比 60/40 更深：273 个滚动十年窗口里它只有 98 个（36%）跑赢。含 11 张图表（其中 1 张可拖杠杆与融资成本）与 2 张表。
The principle — allocate by risk rather than by money — is tested on 392 monthly US observations from 1993-02 to 2025-09: an equal-risk-contribution portfolio of stocks, long bonds, gold and commodities (weights re-solved every January on 36 months of covariance) earns 7.40% a year, with 7.2% volatility, a −22.5% maximum drawdown and a 0.69 Sharpe over 32.7 years. The mechanism holds — 60% of a 60/40's money carries 93.5% of its risk, and risk parity flattens that to 27.3 / 21.7 / 24.5 / 26.6 — but the return does not follow: it gives up 1.3 points a year to a 60/40 and even loses to a naive 25/25/25/25 (0.73 Sharpe against 0.69). Buying those points back requires leverage, and the levered version fell further than a 60/40 in 2022: across 273 rolling ten-year windows it beat the 60/40 in 98 of them, or 36%.
入门Start here组合Portfolio
The 60/40, tested: does the most famous allocation rule hold up?60/40 及其变体：最有名的配置规则经得起检验吗
2026-10-07 · 17 min
用 1990-01 至 2025-09 共 428 个月的美国数据回测 60/40 及其变体。它年化 9.09%、波动 9.1%、最大回撤 −27.0%、夏普 0.70——而 60% 这个数字恰好落在最差的区间：40–50% 的夏普同为 0.72，最大回撤只有 −21.5%。它真正的失效场景是股债同跌：2022 年两条腿各跌 18.1%，全样本有 13.5% 的月份两条腿同时下跌，CPI 高于 5% 时相关性从 −0.04 翻到 +0.38。把 10% 挪给黄金是最便宜的修补（夏普 0.76、回撤 −22.7%，代价是终值少 4.2%）。含 10 张图表（其中 1 张可拖动股票权重）与 2 张附录表。
428 monthly US observations from 1990-01 to 2025-09 test the 60/40 and its variants: 9.09% a year, 9.1% volatility, a −27.0% maximum drawdown and a 0.70 Sharpe. Sixty percent sits in the worst part of the curve — a 40/60 or 50/50 has the same 0.72 Sharpe with a −21.5% drawdown. What breaks it is stocks and bonds falling together: both legs lost 18.1% in 2022, 13.5% of all months saw both fall, and the correlation flips from −0.04 to +0.38 once CPI is above 5%. Shifting 10% into gold is the cheapest repair (Sharpe 0.76, drawdown −22.7%, at a 4.2% cost to terminal wealth).
组合Portfolio信号Signal
Trend following, tested: insurance, not an engine趋势跟踪实测：它是一份保险，不是一台发动机
2026-10-06 · 17 min
用 11 个 ETF、4,931 个交易日检验跨资产时序动量。信号本身有效（方向命中率 53%，30 组参数全为正），但做空那条腿在长牛资产上一直赔钱：多空夏普 0.38、只做多 0.58，都跑不赢 60/40 的 0.75。它真正的价值在坏日子——标普最差 10% 的交易日里跌幅只有 60/40 的八分之一，2022 年 +13.9% 对 −16.6%。含 7 张图表与一个可拖目标波动率的交互动画，配套代码与 chart pack 已开源。
11 ETFs and 4,931 trading days test cross-asset time-series momentum. The signal works (a 53% direction hit rate, all 30 parameter sets positive), but the short leg bleeds in a universe that mostly rises: long-short Sharpe 0.38, long-only 0.58, against 0.75 for a 60/40. Its value is conditional — one eighth of the 60/40's loss on the worst 10% of days, and +13.9% in 2022 against −16.6%.
入门Start here组合Portfolio
The Permanent Portfolio, tested: does an even split cover every environment?永久组合实测：四等分真的能应对任何环境吗
2026-10-05 · 18 min
股、长债、黄金、现金各 25%，用 1990-01 至 2025-09 共 428 个月的美国数据检验。它年化 7.18%、波动 5.8%、最大回撤 −15.5%、夏普 0.75，是四个组合里最高的；25% 的资金对应 37.5 / 33.9 / 28.4 / 0.2 的风险贡献——四等分资金意外地做到了等风险。但 CPI 高于 5% 的 26 个月里它年化 −2.8%，黄金腿 −2.95%，它并没有抗住通胀。含 8 张图表与一个可拖黄金权重、可播放历史的交互动画。
25% each in stocks, long bonds, gold and cash, tested on 428 monthly US observations from 1990-01 to 2025-09: 7.18% a year, 5.8% volatility, a −15.5% maximum drawdown and the highest Sharpe of the four portfolios (0.75). Equal money turns out to be very nearly equal risk (37.5 / 33.9 / 28.4 / 0.2) — but it lost 2.8% a year across the 26 months with CPI above 5%, with gold at −2.95%.
入门Start here组合Portfolio
All Weather, tested: does not forecasting actually survive every cycle?全天候组合实测：不预测，真的能穿越周期吗
2026-10-04 · 16 min
用 1990-01 至 2025-09 共 428 个月的美国数据检验全天候组合的公开描述。它年化 7.6%、波动 7.6%、最大回撤 −22.3%，比 60/40 稳得多（9.2% / −29.6%），但收益更低（8.9%）夏普更差（0.64 对 0.68）。更关键的是它并没有做到环境中性：40% 的长债贡献了 53.5% 的风险，商品腿只有 1.2%。含 6 张图表与 1 个可拖股票权重的交互动画。
428 monthly US observations test the public description of All Weather: 7.6% a year, −22.3% maximum drawdown, steadier than a 60/40 but with lower return and Sharpe. It is not environment-neutral — long bonds carry 53.5% of the risk while commodities carry 1.2%.
组合Portfolio
Can investing beat inflation?投资能跑赢通胀吗
2026-10-04 · 15 min
用 1990-01 至 2026-08 共 432 个月的美国数据检验「投资能跑赢通胀」。股票年化实际收益 8.4%、债券 1.9%、现金只有 0.1%——最初的 100 元 36 年后只值 102 元，而且持有 20 年仍有六成窗口跑输通胀。2022 年通胀 6.5% 时五类资产实际收益全为负。含 6 张图表与 1 个可拖持有期的交互动画。
432 monthly US observations test whether investing beats inflation. Real annualised returns: equities 8.4%, bonds 1.9%, cash just 0.1% — 100 units of cash became 102 after 36 years, and cash still trailed inflation in 59.9% of twenty-year windows. In 2022 all five assets lost in real terms.
组合Portfolio
The ten best days: a statistic used the wrong way错过最好的十天：一句被误用的统计
2026-10-03 · 14 min
用标普 500 从 1970 年至今 14,309 个交易日检验「错过最好的 10 天，收益腰斩」。全仓年化 8.10%，错过最好 10 天降到 6.52%——但躲过最差 10 天升到 10.12%，幅度还更大。随机剔除 10 天则几乎没影响，而最好的日子与最差的日子只隔几天。含 6 张图表与 1 个可拖天数的交互动画。
14,309 trading days test the claim that missing the ten best days halves your return. Buy-and-hold compounds at 8.10% and missing the best ten cuts it to 6.52% — but avoiding the worst ten lifts it to 10.12%, a bigger move. Missing ten random days changes almost nothing, and the best days sit days away from the worst.
入门Start here组合Portfolio
Gold: an inflation hedge, or a volatility hedge?黄金：抗通胀，还是抗波动？
2026-10-03 · 15 min
以 1968-01 至 2026-08 共 703 个月的伦敦金价与美国 CPI 检验「黄金抗通胀」。58 年看它确实跑赢（实际购买力 12.5 倍），但 1980 年买在高点要等 45 年才回本，持有十年有四成窗口是亏的。它真正对冲的是组合波动：与股票相关性 −0.08，加入 10% 可把最大回撤从 28.4% 降到 24.6%。含 7 张图表与 1 个可拖权重的交互动画。
703 monthly observations of the London gold price and US CPI test the claim that gold hedges inflation. Over 58 years real purchasing power rose 12.5-fold — but a January 1980 buyer waited 45 years to break even, and 40% of ten-year windows lost money after inflation.
入门Start here组合Portfolio信号Signal
When VIX tops 30, should you be greedy?VIX 冲上 30 的时候，该贪婪吗？
2026-10-03 · 16 min
「别人恐惧我贪婪」在收益层面站得住：VIX ≥ 30 之后 12 个月的中位收益是 21.8%。但 36 年半里这样的机会只有 7 个互不重叠的市场阶段，其中一次买入后两年仍在亏损。含 8 张图表与 1 个可拖阈值的交互动画。
"Be greedy when others are fearful" survives on returns: the median 12-month return after VIX ≥ 30 is 21.8%. But 36 years produced only seven independent episodes, one of which was still losing money two years later.
组合Portfolio
Dollar-cost averaging vs lump sum: evidence from 942 monthly observations定投与一次性投入：基于 942 个月度观测的实证检验
2026-10-02 · 15 min
研究工作稿：以标普 500 全收益与恒生指数共 942 个月度观测检验一次性投入与分期定额投入（DCA）。一次性投入在约八成起点占优，其劣势几乎全部集中于「部署期下跌」的样本；换到恒生指数，优势降至 63–69%。含 9 张图表与 2 个交互动画，文末附参考文献。
A working paper on DCA versus lump sum using 942 monthly observations across two markets: all-in wins about 80% of starts, and its losses concentrate in falling deployment windows.
组合Portfolio
Every constraint has a price约束的代价：多资产组合里，每个"不许超"都在收费
2026-09-29 · 12 min
同样 8% 波动预算下，四条典型机构约束一共拿走 46.8 bps/年；但样本短的时候，同样的约束是保险而不是成本。四张图 + 八条可点开的文献。
Four ordinary institutional constraints cost 46.8 bps at an identical risk budget — and act as insurance when the sample is short.
信号Signal
Factor screening: why an in-sample Sharpe of 1.95 can be worth zero因子筛选：为什么样本内 1.95 的漂亮结果，样本外会归零
2026-09-25 · 12 min
同一份模拟数据，强弱两档信噪比：弱信号下按 t 值挑出的前 10 名，样本内夏普 1.95、样本外 0.03。附「证明一个因子需要多少年数据」的门槛计算与 11 条可点开的文献。
Two signal-to-noise regimes on the same simulation: with a weak signal, the top ten factors by t-statistic earn 1.95 in sample and 0.03 out of sample.
模型与验证Models & Validation
Glossary: System One, RLCD, calibration, type safety术语表：System One · RLCD · 校准 · 类型安全
2026-09-25 · 5 min
本文里出现的每个术语，给一句话定义、一个「为什么重要」和一条出处。其他笔记直接引用这里的锚点。
Every term used in these notes: a one-line definition, why it matters, and a source anchor other notes link to.