Correlation Must Be Verified

Dear Shareholders,

致股東：

Diversification is often presented as a property of asset labels. We treat it as a state-dependent empirical claim. Gold, equities, government bonds, credit, and commodities do not carry permanent portfolio roles. Their behavior changes with the reason for the shock, the source of financing pressure, and the identity of the marginal seller.

分散投資經常被描述成資產標籤本身的特性。我們則把它視為一項依賴市場狀態、必須持續驗證的實證主張。黃金、股票、政府債券、信用債與商品並沒有永久固定的組合角色。它們的行為會隨衝擊原因、融資壓力來源，以及邊際賣方的身分而改變。

Our trading framework classifies relationships as normal, weakened, or crisis. In a normal regime, historical correlation and expected economic linkage broadly agree. In a weakened regime, the hedge may still work but with reduced efficiency, delayed response, or unstable basis. In a crisis regime, liquidity demand can dominate fundamental narratives and turn several assets into the same trade. Position sizing, hedge ratios, and exit assumptions must change with the state.

我們的交易框架把資產關係分成正常、減弱與危機三種狀態。在正常狀態下，歷史相關性與預期的經濟連結大致一致。在減弱狀態下，避險仍可能發揮作用，但效率降低、反應延遲或基差不穩定。在危機狀態下，流動性需求可能壓過基本面敘事，令多種資產變成同一筆交易。部位大小、避險比率與退出假設都必須隨狀態改變。

Our market thinking focused on a difficult possibility: an asset widely described as a safe haven can fall at the same time as equities. This can happen when real yields rise, the currency strengthens, volatility forces deleveraging, or profitable positions are sold to fund losses elsewhere. The correct question is not whether an asset is defensive in ordinary history. It is whether the asset is held by investors who may become forced sellers under the specific scenario we fear.

本期市場思考集中於一個困難但真實的可能性：被廣泛稱為避險資產的工具，可能與股票同時下跌。當實質利率上升、貨幣轉強、波動迫使去槓桿，或投資人出售獲利部位以支付其他損失時，這種情況便可能發生。正確問題不是某項資產在一般歷史中是否防禦，而是在我們真正擔心的情境下，持有該資產的人是否可能成為被迫賣方。

Our trading action was to reduce gross confidence before reducing analytical effort. We did not replace one unreliable hedge with another narrative-driven hedge. Instead, we combined smaller exposures, higher cash optionality, shorter repricing instruments, and explicit loss limits. We treated cash not as a forecast of falling markets but as an instrument that has no correlation surprise and provides execution capacity during dislocation.

交易行動方面，我們先降低總體信心，而不是降低分析強度。我們沒有用另一個敘事驅動的避險取代原本不可靠的避險，而是結合較小曝險、更高現金選擇權、較短重新定價工具與明確損失限額。我們不把現金視為看空市場的預測，而是把它當成不會產生相關性意外、並能在市場失序時提供執行能力的工具。

The year-to-date position remains a profit of +0.00005 unit. The result is intentionally reported with no rhetorical inflation. A nearly flat outcome can be acceptable when the market has not offered a well-defined asymmetry and when preserving risk capacity is the dominant mandate. Flatness becomes failure only when it reflects indecision, hidden losses, or an inability to identify opportunity. Our flatness reflects a conscious price for waiting.

本年度至今部位仍為 +0.00005 單位正收益。我們刻意不對這個結果作修辭上的放大。當市場尚未提供清楚的不對稱機會，而且保留風險容量是主要任務時，接近持平可以是合理結果。只有在持平源自猶豫、隱藏損失或無法辨識機會時，它才代表失敗。我們的持平，是為等待而支付的有意識成本。

The lesson learned is that correlation matrices built from calm periods can create false precision. A single average obscures nonlinear behavior, lead-lag effects, and the difference between daily noise and multiweek stress. We also learned that a hedge with poor liquidity may be theoretically correct and practically unusable.

本期教訓是，利用平靜期間建立的相關性矩陣容易製造虛假的精確感。單一平均值會掩蓋非線性行為、領先落後關係，以及每日雜訊與數週壓力之間的差別。我們也確認，流動性差的避險可以在理論上正確，卻在實務上無法使用。

Correlation is reviewed across daily, weekly, and rolling ninety-day windows because different horizons answer different questions. Daily correlation helps identify immediate liquidation behavior. Weekly correlation reveals whether a shock is propagating beyond noise. The ninety-day window helps determine whether the portfolio has entered a new operating regime. We also inspect conditional correlation during the worst market observations rather than relying only on the full-sample average.

我們會同時檢視每日、每週與九十天滾動相關性，因為不同時間尺度回答不同問題。每日相關性有助辨識即時清算行為；每週相關性顯示衝擊是否已超越雜訊並持續擴散；九十天窗口則協助判斷投資組合是否進入新的運作狀態。我們也會檢視市場最差觀察期間的條件式相關性，而不是只依賴全樣本平均值。

Crowding can make correlation unstable even when fundamental exposures differ. Two assets may have unrelated cash flows yet share the same leveraged owners, volatility controls, collateral channels, or redemption-sensitive funds. When these investors reduce risk simultaneously, ownership structure temporarily becomes more important than economic classification. Our liquidity map therefore includes market depth, dealer balance-sheet capacity, ownership concentration, ETF activity, derivatives positioning, and the speed at which quoted liquidity disappears under stress.

即使兩項資產的基本面曝險不同，擁擠也可能令相關性不穩定。兩項資產的現金流可以互不相關，卻由相同的槓桿投資人、波動控制策略、抵押品渠道或贖回敏感型基金持有。當這些投資人同時降低風險時，持有人結構會暫時比經濟分類更重要。因此，我們的流動性地圖會涵蓋市場深度、交易商資產負債表能力、持有人集中度、ETF 活動、衍生品倉位，以及報價流動性在壓力下消失的速度。

A hedge is evaluated by payoff timing as well as payoff direction. Protection that arrives after the portfolio has already breached its liquidity or drawdown limit is not effective protection. We therefore estimate how quickly the hedge reprices, whether it can be monetized during stress, and whether gains are likely to be offset by widening basis, higher margin, or deteriorating counterparty terms.

避險不只要依照報酬方向評估，也要依照報酬出現的時點評估。若保護在投資組合已突破流動性或回撤限制後才出現，它就不是有效保護。因此，我們會估計避險重新定價的速度、壓力期間能否變現，以及收益是否可能被基差擴大、保證金提高或交易對手條件惡化所抵銷。

We distinguish economic hedges from accounting hedges. An economic hedge reduces the underlying sensitivity even if short-term reported P&L remains noisy. An accounting hedge may smooth reported variation while leaving the fundamental exposure largely unchanged. Our priority is economic loss containment. Reporting stability matters, but it must not create an incentive to retain risks that remain economically open.

我們區分經濟避險與會計避險。經濟避險會降低底層敏感度，即使短期報告損益仍有雜訊；會計避險可能平滑報告波動，卻讓基本曝險大致維持不變。我們優先控制經濟損失。報告穩定性固然重要，但不能因此誘使投資組合保留實際上仍然開放的風險。

Hedge sizing is based on stressed sensitivity rather than a single historical beta. We consider the exposure if volatility doubles, correlations converge, liquidity haircuts increase, and the hedge performs only partially. This produces a more conservative hedge ratio and makes residual risk visible. Residual risk is not automatically a defect, but it must be intentional, measured, and supported by sufficient capital.

避險規模以壓力敏感度為基礎，而不是使用單一歷史 beta。我們會考慮波動加倍、相關性收斂、流動性折價提高，以及避險只部分生效時的曝險。這會產生較保守的避險比率，並令剩餘風險更加可見。剩餘風險不一定是缺陷，但它必須是有意識、可衡量，而且有足夠資本支持的結果。

Our next action is to maintain a live correlation map across rolling horizons and assign every hedge a documented failure mode. We will not let the map become a substitute for causal reasoning. Statistical evidence will be paired with a funding map that identifies who owns the asset, how it is financed, what collateral rules apply, and where forced selling may emerge. For each hedge, we will state whether likely failure could arise from basis risk, delayed response, liquidity loss, excessive carry cost, option decay, counterparty exposure, or a reversal in the underlying economic relationship. New hedges will be approved only when both the economic mechanism and the execution mechanism are credible, because a hedge without a documented failure mode creates false confidence precisely when discipline is most needed.

下一步，我們會在不同滾動週期下維護即時相關性地圖，並為每項避險指定及記錄失效模式。我們不會讓統計地圖取代因果推理，而會把統計證據與融資地圖配對，檢視誰持有資產、如何融資、適用哪些抵押規則，以及強迫賣出可能在哪裡出現。對每項避險，我們會說明最可能的失效是否來自基差風險、反應延遲、流動性流失、持有成本過高、選擇權時間價值衰減、交易對手曝險，或底層經濟關係逆轉。只有當經濟機制與執行機制都可信時，新的避險才會獲得批准，因為沒有記錄失效模式的避險，往往會在最需要紀律時製造虛假信心。