全天候组合实测：不预测，真的能穿越周期吗All Weather, tested: does not forecasting actually survive every cycle?

这张图不是回测，是它的设计说明——先看清主张，才谈得上检验。关键在于分母是风险不是资金：债券的波动只有股票的四分之一左右，所以要让四个格子风险对等，要么让债券的资金权重远大于股票，要么整体加杠杆。这也解释了为什么机构版需要借钱，而大众版只能靠堆债券权重来近似。

This is not a backtest but the design statement — the claim has to be stated before it can be tested. The denominator is risk, not capital: because bonds are roughly a quarter as volatile as equities, equalising risk across environments requires either far larger bond weights or leverage.

来源：根据全天候的公开描述整理。Source: compiled from the public description of the strategy.

按全样本协方差分解，40% 的长债贡献了 53.5% 的组合风险，15% 的中债贡献 10.5%，而两支各 7.5% 的通胀资产——黄金 2.4%、商品 1.2%——加起来不到 4%。「四个环境各占 25% 风险」这件事，在大众版里并没有发生。原因很直接：没有杠杆时，你无法把债券的风险抬到股票的水平，只能靠加大它的资金权重，而这样做出来的组合，风险其实集中在一两支资产上。

By full-sample variance decomposition, a 40% long-bond weight contributes 53.5% of portfolio risk, while the two inflation assets — gold at 2.4% and commodities at 1.2% — sum to under 4%. The "25% of risk per environment" simply does not happen in the unlevered version.

来源：本文计算，全样本协方差。Source: author's calculations, full-sample covariance.

对数刻度，所以斜率就是年化增速。全天候（红）的线最平——它确实更稳；但 60/40（蓝）在它上方，纯股在最上方。35 年下来，1 元变成约 13.7 元（全天候）、21.0 元（60/40）、41.9 元（纯股）。这不一定说明全天候不好——它的风险更低——但足以说明一件事：「全天候」这个名字容易被读成"收益也不错"，而它在数据里并不是。

On a log scale the slope is the growth rate. All Weather is the flattest line — genuinely steadier — but the 60/40 sits above it and equities above that.

来源：本文回测，年度再平衡，不扣交易成本。Source: author's backtest, annual rebalancing, before transaction costs.

左图里全天候落在最左下——波动 7.6% 换成 7.6% 的年化；60/40 在它右上方，波动 9.2% 换 8.9%。右图是它真正的优势：最大回撤 −22.3%，而 60/40 是 −29.6%、纯股是 −50.9%。所以准确的表述是：它用 1.3 个百分点的年化，买了 7 个百分点的回撤改善。夏普上这笔交易并不划算（0.64 对 0.68），但对一个受不了深回撤的投资者，它可能仍然值得。

All Weather sits bottom-left; the 60/40 is up and to the right. Its real edge is the drawdown: −22.3% against −29.6% and −50.9%. In Sharpe terms the trade is not favourable (0.64 versus 0.68) — but for someone who cannot sit through a deep drawdown, it may still be worth it.

来源：本文回测；夏普以 3 个月国库券为无风险利率。Source: author's backtest; Sharpe uses three-month T-bills as the risk-free rate.

全天候在四格里的年化是 6.2 / 8.0 / 9.7 / 7.6，最高与最低差 1.6 倍；60/40 是 7.9 / 10.9 / 10.0 / 6.0，差 1.8 倍——全天候确实更均衡一些，但没有质的变化。真正的差别在结构上：它只在「增长下降·通胀下降」领先 60/40 1.6 个百分点，「增长下降·通胀上升」基本持平（−0.3），两个「增长上升」的格子分别落后 1.7 与 2.9 个百分点。也就是说，它把注押在了"增长转弱"上——那是债券真正发挥作用的环境，也正好对应它 53.5% 的风险集中在长债这件事。

All Weather ranges from 6.2% to 9.7% across the four cells; the 60/40 from 6.0% to 10.9%. It is somewhat more balanced but not qualitatively so. The structural point: it leads the 60/40 in only one of the four cells (+1.6 points) and trails by 0.3, 1.7 and 2.9 points in the other three — the opposite of "neutral".

来源：本文计算；环境划分不使用资产价格。Source: author's calculations; the classification uses no asset prices.

左图是各条腿对年化的算术贡献：股票 3.5 个百分点（资金 30%）、长债 2.6（资金 40%）、中债 0.8、黄金 0.5、商品 0.3。30% 的股票贡献了最多收益，40% 的长债贡献了最多风险——这就是大众版的真实结构。右图是加杠杆的结果：1.5 倍的年化升到 9.6%，但最大回撤扩到 −33.1%，已经比 60/40 的 −29.6% 更深；2 倍时回撤达到 −43.8%。把长债换成现金（反事实）年化从 7.6% 掉到 6.3%——久期既是它的收益来源，也是它的风险来源，两者是同一件事。

Attribution: equities contribute 3.5 points of return on a 30% weight; long bonds 2.6 points on 40% — and 53.5% of the risk. Leverage at 1.5x lifts the return to 9.6% but the drawdown to −33.1%, deeper than the 60/40. Replacing long bonds with cash cuts the return to 6.3%: duration is the return source and the risk source at once.

来源：本文计算；融资成本按 3 个月国库券 + 50bp。Source: author's calculations; financing at three-month T-bills plus 50bp.

把股票权重从 30% 往上拖，其余四条腿按原版比例缩放（也就是保持"债券为主"的结构）。可以清楚看到一条规律：年化和回撤几乎同步上升——拖到 60%，收益接近 60/40，但回撤也追上了它；拖到 90%，就更像纯股。这个交互想说的不是"哪个权重最好"，而是：全天候给出的那条"低回撤"曲线，不是免费的，它的价格就写在收益那一栏里。

Drag the equity weight up; the other four legs scale proportionally. Return and drawdown rise almost together — at 60% you get close to a 60/40's return and its drawdown. The point is not which weight is best, but that All Weather's smoother path has a price, and the price is in the return column.

来源：本文回测，年度再平衡，不扣交易成本。Source: author's backtest, annual rebalancing, before costs.