A macro signal that changes its mind four times in six weeks has not learned anything new. It has parked next to a line. Almost every complaint I hear about liquidity indicators being unreliable turns out on inspection to be this one mechanical situation, and once you can see it you stop taking those flips personally and you stop trading them.
A continuous score meeting a hard line
The Global Liquidity Scorecard produces two kinds of output at once. There is a continuous number, the composite, which currently reads 85 on a 0 to 100 scale. And there are categorical labels, REGIME reading RISK-ON and POLICY reading EASING. The labels are what most people act on, and labels are made by drawing a line through a continuous number.
That is the whole mechanism. Wherever the boundary sits, and I do not know where it sits because the module shows me the labels rather than the cut points, a categorical output changes state the instant the underlying number crosses it. The size of the crossing is irrelevant to the label. A move from one side to the other by a tenth of a point produces exactly the same change of state as a move of thirty points.
So the behaviour of the label depends entirely on where the composite is sitting relative to that line. Out at 85, comfortably inside a band, week-to-week wobble of a couple of points does nothing at all and the label is stable for months. Sitting a point away from a boundary, the same wobble flips the label every time it happens. Same indicator, same data quality, completely different experience, and the only variable that changed is distance to the line.
Where the wobble actually comes from
The flips are not random noise in the abstract. They come from four identifiable places, and knowing which one is moving tells you whether to care.
Currency translation is the biggest. Eight central banks report in their own currencies, and the BoJ, PBoC, ECB, BoE, SNB, BoC and RBA all have to be expressed in a common unit before they can be added to the Fed. A two percent move in the dollar restates a large slice of the aggregate without a single policy decision anywhere in the world. For a composite sitting near a band edge, a routine currency week is enough to flip the label on its own.
Second is the release calendar. The inputs publish weekly and monthly, and they do not publish together. A week where three of the eight report and five do not gives you a composite built partly from fresh data and partly from stale data, and the mix changes every week. That produces a sawtooth pattern in any aggregate, entirely independent of what conditions are doing.
Third is revision. Monetary aggregates get restated. A number you acted on can quietly become a different number later, which also means that when you look back at a chart of the composite it will be smoother than the series you actually lived through.
Fourth is credit spreads, which are a market price rather than a reported statistic and therefore move daily. Mixing a daily-moving input with weekly and monthly ones means the composite has a permanent low-level jitter floor that has nothing to do with policy.
None of these four is a defect. They are what happens when you combine eight national accounting series and a market price into one figure. But they do mean the composite has an irreducible wobble of a couple of points, and any decision rule that cannot survive that wobble will churn.
The dead zone you build in your own log
I want to be clear about where this rule lives. Trade Signals is a view inside GLS, alongside Dashboard, Regime, Risk, Countries, Data and Trade Analysis. In the capture above it shows the header, the composite and the two labels. I do not see a hysteresis setting, a confirmation control, or a sensitivity slider, and I am not going to claim one exists. The dead zone is something you impose in your own weekly log, on your side of the screen, and it costs nothing but the discipline to keep it.
It has two parts and you need both.
- A buffer. Pick your own action line, then require the composite to be at least a few points past it before you act. Wider going in than coming out is a reasonable asymmetry if you care more about drawdowns than about missing upside.
- A confirmation count. Require the reading to hold on the far side for two consecutive weekly checks before you move any money. One reading is an observation. Two is a state change.
Take an invented but entirely ordinary sequence of weekly readings against an action line you have set at 70: 71, 69, 72, 68, 71. Acting on the raw label, that is four exposure changes in five weeks. With a three-point buffer and a two-reading confirmation, it is zero changes, because nothing ever gets three points clear of the line and stays there for a fortnight. Both rules read the same data. One of them made you trade four times to end up exactly where you started.
What the dead zone costs you
Every filter is a trade of one error for another, and it is worth pricing the trade rather than assuming the quiet version is free.
The saving first. On a 40,000 dollar account where an exposure change moves 22,000 dollars, four avoidable round trips at ten basis points of spread and slippage costs you roughly ninety dollars. That is real but it is not the main number. In a taxable account the bigger cost of churn is that each flip realises short-term gains you had no reason to realise, converting a deferral into a bill at your marginal rate. And the largest cost is the one that never appears on a statement, which is that a rule you have watched fail four times in a row is a rule you will stop following in month three, precisely when it finally matters.
Now the cost. A two-reading confirmation on weekly data means you are up to two weeks late on every genuine turn, in both directions. If a real deterioration begins and the index falls six percent over those two weeks, the delay on that same 22,000 dollars of exposure costs about 1,300 dollars. That is roughly fourteen times the trading friction you saved. The dead zone is not a free lunch, it is a bet that false turns are more frequent than fast real ones.
Which is usually the right bet with this particular input, because liquidity conditions genuinely do move slowly. Balance sheets do not reverse in a fortnight. The cases where the bet loses are the ones where a fast repricing event drags credit spreads violently in a few days and the composite falls through your buffer while you are waiting for confirmation. That happens, it is exactly when you least want to be two weeks late, and no amount of filter tuning fixes it, because the filter is what made you slow on purpose.
So carry a separate rule for that case rather than pretending the dead zone handles it. Mine is a single override: if the composite moves more in one weekly reading than it usually moves in a quarter, I act on the first reading and skip the confirmation entirely. A large fast move is not the kind of noise the dead zone was built to absorb, and treating it as such is how a sensible filter turns into a reason you did nothing.