Flow-pause monitor

Status history

Conditions

Timeline of changes

Caveats

• Tax receipts timing. The withheld-tax series comes from the Treasury's daily statement. Receipts swing with the number of paydays, public holidays and deposit timing in each 20-day window, so a single print can jump several points (this condition moved from +9.0% to about −0.2% y/y in a few days). Treat a fresh flip as provisional until it holds for a couple of weeks.
• A heuristic, not a forecast. The thresholds were set once, up front, and are judgement calls. RED says several pressures on passive inflows line up at once; it has not reliably timed market tops.
• Inputs start at different times (claims and Cboe history earlier, SPY creations and withheld tax from 2006, TQQQ from 2010), so the count's history is not like-for-like and earlier periods have fewer inputs.
• No look-ahead, but lags differ. Percentiles and averages use only data available at each date. Equity issuance is quarterly and is shifted about 10 weeks for its publication lag; CPI is shifted about two weeks.
• ETF flows are estimates from the change in shares on issue times the prior day's NAV; issuer revisions or share-class changes can move them.
• Single-stock leveraged ETFs are tracked but not counted (n/a, not in the denominator) until weekly snapshots build up.
• Backtest notes: 2007-09 GFC (S&P 500 peak 2007 9 Oct): inputs available: 3/7; count in 12m before peak: max 1, avg 0.5, at peak 0 (prior 4 years avg 0.5); share of days each was on: claims 0%, spy 49%, iss 0% | 2020 Covid crash (peak 2020 19 Feb): inputs available: 6/7; count in 12m before peak: max 3, avg 1.7, at peak 0 (prior 4 years avg 1.4); share of days each was on: claims 8%, withheld 7%, spy 58%, tqqq 98%, corr 0%, iss 0% | 2022 bear market (peak 2022 3 Jan): inputs available: 6/7; count in 12m before peak: max 3, avg 1.8, at peak 1 (prior 4 years avg 1.9); share of days each was on: claims 19%, withheld 2%, spy 33%, tqqq 46%, corr 0%, iss 80% | Base rate since all/most inputs exist: since Feb 2010 the count was >= 3 on 18% of days and >= 4 on 3% of days