1導讀華爾街“超級大腦”集體拉響警報:摩根大通、高盛、花旗、IMF 等掌門罕見同頻,直指股債金“三高”——標普500周期調整市盈率40倍,僅次網際網路泡沫;投資級債信用利差僅0.8%,創2005年最窄;黃金兩日閃崩7%。歷史顯示,低波動後必接高波動“跳檔”,量化模型卻難捕捉疫情、擠兌等“純粹衝擊”。橋水用機器學習掃描萬組宏觀變數,仍難精準擇時。交易員只能把多模型、動量、跨資產相關性當“煙感器”,早一秒止損,卻無法給出崩盤日曆:泡沫已現,雷管未知,現金為王或是最優解。2地道表達territory /ˈterətri/ - n.領土;領域,管區,地盤bubble territory /ˈbʌbl ˈterətri/ – n. 泡沫區間cyclically adjusted earnings – n. 周期調整盈利(CAPE)credit spread – n. 信用利差one’s voice carries - 某人的聲音“傳得遠”。表示某人說話有份量,某人發聲能引起關注[例]:Her voice carries in the industry because she has successfully launched three unicorns. 她在業界說話有份量,因為她已成功創辦了三家獨角獸公司。come down to - “歸根結底在於……;說到底就是……”[原文]:Forecasting a correction comes down to forecasting volatility. 預測回呼歸根結底就是預測波動率。state-of-the-art approach - “最尖端的方法;當前技術最高水平的做法”[例]:The paper presents a state-of-the-art approach to forecasting market volatility using transformer networks. 該論文提出了一種基於 Transformer 網路預測市場波動的最前沿方法。a run on a bank - 銀行擠兌[例]:Governments often provide deposit insurance to prevent a run on banks during financial crises. 政府常在金融危機期間提供存款保險,以防出現銀行擠兌。3|原文|翻譯|字詞|Before the fall崩盤前夕Wall Street won't forecast the next crash華爾街無法預測下一次崩盤A lot of assets, warned Jamie Dimon in mid-October, “look like they’re entering bubble territory”. His voice carries because he runs America’s biggest bank, JPMorgan Chase, but also because it is part of a growing chorus. David Solomon, Mr Dimon’s opposite number at Goldman Sachs, talks of “investor exuberance”; Jane Fraser, Citigroup’s boss, of “valuation frothiness”. The Bank of England recently cautioned that “the risk of a sharp market correction has increased.” The IMF worries about a “disorderly” one, since “risk asset prices are well above fundamentals”.10月中旬,傑米·戴蒙警告說:“大量資產看起來正在進入泡沫區間。” 他的話之所以有份量,不僅因為他執掌美國最大的銀行——摩根大通,也因為越來越多的重量級人物正同聲附和。高盛的“對應角色”戴維·所羅門談及“投資者狂熱”;花旗集團老闆簡·弗雷澤則提到“估值泡沫化”。英國央行最近提醒,“市場急劇回呼的風險已上升”。國際貨幣基金組織擔心出現“無序”回呼,因為“風險資產價格已遠高於基本面”。They are all right about one thing: valuations of many assets are dizzyingly high. To buy the basket of stocks in America’s S&P 500 share index, investors now pay 40 times their underlying, cyclically adjusted earnings—a multiple exceeded only during the dotcom bubble, and then not by much. Investment-grade corporate bonds, on average, yield only 0.8 percentage points more than equivalen Treasuries. The last time this credit spread was so low was in 2005, during the run-up to the global financial crisis. Even gold, normally a haven, seems vulnerable to traders’ whims. After soaring to a record high on October 20th its price dropped by 7% in just two days, and is now 9% below its peak.他們有一點說得沒錯:許多資產的估值已高得令人眩暈。如今要買入代表美國股市的標普500指數一籃子股票,投資者需要支付其周期調整盈利(CAPE)的40倍——這一倍數僅在網際網路泡沫時期被超越過,且當時也沒高出多少。投資級公司債券平均只比同期美國國債多提供0.8個百分點的收益率;這種信用利差如此之低,還要追溯到2005年全球金融危機暴發前的醞釀期。就連通常被視為避風港的黃金,眼下也顯得易受交易員一時興起的影響:10月20日剛飆升至歷史新高,短短兩天內就跌去7%,目前比峰值低9%。Perhaps, then, one correction has already begun. When might others follow? For anyone who is hoping to beat the market—including the quant shops and other hedge funds that promise returns whatever the weather—there are few more pressing questions. “The nirvana of macro trading is forecasting these turning-points,” says James White of Elm Wealth, an investment firm. The snag is that achieving nirvana is just about impossible.或許,一輪迴調已然啟動;下一輪又會何時到來?對於任何想跑贏市場的人——包括那些號稱“全天候”都能賺錢的量化基金和避險基金——幾乎找不到比這更緊迫的問題。投資公司Elm Wealth的詹姆斯·懷特說:“宏觀交易的理想狀態就是預判這些拐點。” 可難點在於,想要真正達到這種理想狀態幾乎是不可能的。territory /ˈterətri/ - n.領土;領域,管區,地盤bubble territory /ˈbʌbl ˈterətri/ – n. 泡沫區間voice carries – 聲音有份量;說話有人聽exuberance /ɪɡˈzjuːbərəns/ – n. 狂熱;過度興奮frothiness /ˈfrɒθinəs/ – n. 泡沫化;虛高dizzyingly /ˈdɪzɪɪŋli/ – adv. 令人眩暈地;極高地cyclically /ˈsaɪklɪkəli/ - 循環的;周期性地vulnerable /ˈvʌlnərəbl/ – adj. 易受影響的whim /wɪm/ - n.心血來潮;一時的興致nirvana /nɪəˈvɑːnə/ – n. 涅槃;理想境界,終極追求snag /snæɡ/ – n. 小障礙;棘手之處Forecasting a correction comes down to forecasting volatility (or “vol”): the propensity of asset prices to jump around. The traders and quants who attempt this with stocks rely on a set of stylised facts, which become apparent if you stare at a chart of a share-price index for long enough. One is that volatility is “bimodal”, meaning prices spend long periods moving by relatively small amounts each day, interspersed with much shorter ones during which they move a lot (see chart). Volatility, in other words, has a tendency to jump itself: there are low-vol regimes and high-vol regimes, with not much in between. These regimes are correlated with the broad direction asset prices are moving in. During low-vol periods, prices tend to be stable or rising; during high-vol regimes, they tend to fall.歸根結底,預測回呼就是預測波動率(行內簡稱“vol”):即資產價格上下跳動的傾向。試圖對股票做此預測的交易員與量化分析師,會依賴一套“程式化事實”——只要你把股價指數圖盯得夠久,這些規律就會浮現。其中之一是波動率呈“雙峰”:價格在很長時期內每天變動很小,中間卻穿插著短暫的大幅震盪。換言之,波動率本身也會“跳檔”:市場要麼處於低波動狀態,要麼處於高波動狀態,幾乎沒有中間地帶。這兩種狀態與資產價格的總體方向高度相關:低波動時期價格往往平穩或上漲;高波動時期則多半下跌。The fact that volatility mostly stays in one regime or another means that, most of the time, the best way to predict tomorrow’s vol is from today’s. This is the basis for the “autoregressive” models used by traders at banks and investment firms to gauge risk. Such models are also used to price options—contracts that allow traders to profit from big swings in prices—since this requires an estimate of expected volatility. Yet though useful from day to day, these models are ill-equipped to predict the sudden vol jumps that accompany corrections, when the regime changes.波動率往往要麼維持低狀態、要麼維持高狀態,這一事實意味著:絕大多數時候,預測明天波動率的最佳依據就是今天的波動率。這正是銀行和投行的交易員用來衡量風險的“自回歸”模型的基礎。同樣的模型也被用來為期權定價——期權是讓交易員能從價格大幅波動中獲利的合約,而定價需要對預期波動率進行估算。然而,這些模型雖然日常有用,卻難以應對伴隨回呼而來的“狀態切換”時刻,即波動率驟然跳升的情形。As a result, many traders build other models to complement their autoregressive ones, based on outside factors that might cause volatility to leap. The state-of-the-art approach is to use machine learning to analyse the impact of every economic variable under the sun. Corporate earnings, GDP, inflation, jobs data—any or all might contribute to a sudden dip in sentiment. Machine learning, meanwhile, is an ideal tool for spotting combinations that might indicate vulnerabilities which are not intuitively obvious. Bridgewater, one of the world’s biggest hedge funds, is famous for trading strategies based on intricate combinations of economic variables.因此,許多交易員會搭建其他模型來補充自回歸模型,這些模型基於可能導致波動率驟升的外部因素。最尖端的做法是利用機器學習,分析“天底下”所有經濟變數可能帶來的影響——公司盈利、GDP、通膨、就業資料,任何一項或全部都可能突然拉低市場情緒。與此同時,機器學習擅長髮現那些非直觀可見、卻可能預示脆弱點的變數組合。全球最大避險基金之一橋水(Bridgewater)就以基於複雜經濟變數組合的交易策略而聞名。propensity /prəˈpensəti/ – n. 傾向,習性stylise /ˈstaɪlaɪz/ - v. 使……程式化;把……規範化bimodal /baɪˈməʊdl/ – adj. 雙峰的intersperse /ˌɪntəˈspɜːs/ – v. 散佈,點綴,穿插regime /reɪˈʒiːm/ – n.(技術)狀態,體制autoregressive /ˌɔːtəʊrɪˈɡresɪv/ – adj. 自回歸的(統計模型)gauge /ɡeɪdʒ/ – v. 衡量,評估ill-equipped /ˌɪl ɪˈkwɪpt/ – adj. 能力不足的,難以勝任的accompany /əˈkʌmpəni/ - vt.伴隨;陪同;陪伴;與...同時發生leap /liːp/ – v. 驟升,跳躍dip /dɪp/ – n. 下降;(情緒)回落vulnerability /ˌvʌlnərəˈbɪləti/ – n. 脆弱點intricate /ˈɪntrɪkət/ – adj. 錯綜複雜的Unfortunately, few such strategies actually seem to work well enough to beat the market. “I know of no other firm than Bridgewater that has been successful with these macro models,” says a former trader at a rival hedge fund. And even the best of them cannot be expected to predict a “pure shock”, like a pandemic or a run on a bank. The very unpredictability of such shocks means they have the greatest potential to trigger a correction. Macro models can pinpoint situations in which one might have an outsize impact on markets, yet not when it will occur.遺憾的是,這類策略似乎極少能真正跑贏市場。“據我所知,除了橋水,沒有那家避險基金能把這些宏觀模型用成功。”一家 rival 避險基金的離職交易員說道。即便最頂尖的模型,也無法預判“純粹衝擊(不可預見的黑天鵝事件)”——例如一場疫情或銀行擠兌。正是這類衝擊的不可預測性,使它們最有可能觸發市場回呼。宏觀模型可以精準鎖定某些可能產生巨大影響的局面,卻無從告知它何時爆發。Traders do have other tools that they can use to spot a reversal. Since volatility tends to jump at times when prices change direction, momentum traders often dump their positions immediately after the first big move down snaps a winning streak. Correlations between different asset classes also sometimes provide an early warning that a correction is afoot. The classic sign that investors have switched to “risk-off” mode, for instance, is that share prices fall at the same time as those of havens such as gold and Treasuries rise.交易者確實還有其他工具可以用來識別價格反轉。由於價格掉頭時波動率往往驟升,動量交易員常在首輪大跌打破連漲紀錄後立即清倉。不同資產類別之間的相關性有時也能提前預示回呼正在醞釀。最典型的“避險”訊號,便是股價下跌的同時,黃金、美債等避風港資產價格上漲。A good way of improving on the predictions of flawed models, says the head of quantitative strategies at another big hedge fund, is to combine many independent ones. Even then, the best a trader who lacks divine inspiration can really hope for is to identify a correction early and cut their losses. Wall Street’s finest might warn of a crash—do not expect them to say when it will arrive.另一家大型避險基金的量化策略主管說,改進有缺陷模型預測的好辦法,是把眾多彼此獨立的模型組合起來。即便如此,缺乏“神靈啟示”的交易員最多也只能指望及早識別回呼並止損。華爾街最頂尖的高手或許會警告崩盤將至,但別指望他們說出具體時間。pinpoint /ˈpɪnpɔɪnt/ – v. 精準定位momentum /məˈmentəm/ - n.勢頭;動量;動力streak /striːk/ - n.(尤指體育比賽或賭博中)順的時候,背的時候,運氣,手氣afoot /əˈfʊt/ – adj.(計畫、事件)醞釀中,進行中flawed /flɔːd/ – adj. 有缺陷的divine /dɪˈvaɪn/ - adj.神聖的;神的;上帝的 (Chris的外刊閱讀筆記)