A composite liquidity score of 85 on a 0 to 100 scale can mean two incompatible things. It can mean the stock of liquidity is large relative to its own history, which is a statement about where we are. Or it can mean liquidity is being added quickly, which is a statement about what is happening. Allocators routinely read the first and act on the second, and the gap between those two readings is where a lot of badly framed macro overlays come from.
The Global Liquidity Scorecard does not publish its weighting scheme on the panel. What it publishes is the input list, the aggregate central bank balance sheet, global M2 money supply, USD liquidity indicators and credit spreads across eight central banks, and the resulting composite on a bounded scale, with a regime label and a policy stance beside it. That is enough to work out the answer empirically, and working it out is worth an afternoon, because everything downstream of the score inherits the distinction.
Why the distinction is not academic
The stock of liquidity and its rate of change have different statistical personalities, and they behave differently in exactly the situations where you need a macro overlay to earn its keep.
A level series trends. It spends long stretches high, long stretches low, and it does not mean revert on any horizon a fund cares about. A signal built on the level will therefore be persistent, will produce few state changes, and will spend most of a multi year expansion telling you the same thing. That is fine for strategic tilts and useless for anything with a quarterly review.
A change series oscillates. It crosses zero regularly, it is noisier, and it turns before the level does, by construction, since a level cannot peak until its rate of change has already gone to zero. A signal built on change will fire more often, will be wrong more often in the short run, and will be earlier at every genuine inflection.
The general view in macro, and it is a long standing one rather than any single result, is that risk assets are more responsive to the impulse than to the standing stock. Money that has already been created and is already deployed is priced in. Money arriving is not. If you accept that framing at all, then a level score is a description of conditions and a change score is closer to a signal, and it matters a great deal which one is sitting behind an 85.
The diagnostic that separates them
You do not need the source code. A bounded score reveals its own construction through its behaviour, and there are three tells.
Behaviour during a long, flat expansion. Take a stretch where the underlying stock is high and no longer growing. A level based score sits pinned near the top of its range and refuses to move. A change based score drifts back toward the middle, because the impulse has gone to zero even though the stock has not fallen. This is the single cleanest test and it usually settles the question on its own.
Distribution of the score itself. Log the composite daily for a quarter and look at the histogram. Level based composites cluster at the extremes and cross the middle rarely. Change based composites pile up around the centre with fat behaviour at the edges. A score that has spent thirty consecutive readings above 80 is very unlikely to be measuring a first difference.
Reaction to a publication. When a slow input publishes, watch the size of the step. A change based construction reacts to the difference between the new print and the old one, so a modest revision can move it noticeably. A level based construction absorbs the same print with a much smaller move, because one month of growth barely shifts a stock measured against years of history.
Running the test properly
The version I would actually document looks like this, and it produces something you can hand to a due diligence reviewer rather than an impression.
- Capture the composite on a fixed schedule and store it with the recompute timestamp. One reading per day, same time, no exceptions, because an irregular capture schedule will contaminate everything that follows.
- Build two reference series from the public source data behind the same input families. One is the level of an aggregate balance sheet measure in a fixed currency. The other is its change over a fixed lookback, thirteen weeks is a reasonable default because it is long enough to survive weekly noise and short enough to turn.
- Rank correlate the composite against each reference series over your capture window. Use rank rather than linear correlation, since a bounded 0 to 100 score is almost certainly a monotone transform of something and you care about ordering, not scale.
- Repeat at lags. Shift the composite forward and back by up to eight weeks and see where the correlation peaks. A level score will correlate contemporaneously with the level. A change score will lead the level and correlate contemporaneously with the change.
- Record the answer with the window it was measured over, and re-run it annually. Composites get re-specified, and a construction test from two years ago is a claim about a version that may no longer exist.
Two honest caveats on this procedure. First, the mixed publication frequencies underneath will blur the lag structure, so expect a broad correlation peak rather than a sharp one. Second, the answer may be neither, because a sensible composite can perfectly well blend a level term and a change term, in which case what you will find is meaningful correlation with both and a lead over the level that is shorter than a pure change series would give you. That is a legitimate result and it is more informative than either clean answer.
What changes in the investment process once you know
If the score turns out to track the level, then it is a conditioning variable, not a trigger. It belongs in the part of the process that sets a strategic range, and it should never be the thing that moves a book. You would also stop describing a high reading as an easing impulse in any written document, because an 85 that sits alongside a policy tile reading EASING is not confirmation, it is two different measurements that happen to be pointed the same way today.
If it tracks the change, the obligation reverses. Now it is timely enough to inform positioning, and now the discipline you need is around whipsaw: a change based score will hand you more state transitions than you can responsibly act on, and a rule for how many you ignore has to be written down before the first one arrives, not after.
Either way, the sentence that goes in the memo becomes specific. Not "global liquidity is supportive at 85", which survives no follow up question at all, but "the composite reads 85, which on our own construction test tracks the standing level rather than the impulse, so we are treating it as a permissive condition and not as a reason to add". That sentence tells a reviewer what you believed, why, and what would have changed your mind, which is the entire job of the paragraph.