The Global Liquidity Scorecard states its coverage plainly. Eight central banks, being the Fed, the ECB, the Bank of Japan, the People's Bank of China, the Bank of England, the Swiss National Bank, the Bank of Canada and the Reserve Bank of Australia. Four of those are large enough that their balance sheets dominate any sensibly weighted aggregate. The other four are not, and the question worth an afternoon of research time is whether the second four are carrying information or carrying the appearance of comprehensiveness.
Why the question is not pedantic
Three reasons it matters at desk level rather than as a curiosity.
The first is attribution. If your monthly macro memo explains a composite move by pointing at a policy change at one of the smaller institutions, and that institution's weight cannot plausibly account for the move, then your memo contains a causal claim that is arithmetically impossible. Somebody will eventually check.
The second is data cost and revision risk. Every additional national series in an aggregate brings its own publication lag, its own revision behaviour and its own currency conversion. Those costs are real and they are paid in the tails, when a revision to a small series moves the aggregate at exactly the moment you are relying on it.
The third is that comprehensiveness sells. Eight central banks is a better line in a product description than four, and that incentive exists independently of whether the extra four improve the signal. That is not an accusation about any particular vendor. It is a reason to test rather than to assume.
The three ways a small input can earn its place
Before designing a test, be precise about what would count as passing, because "does it matter" is not a testable statement.
Contribution to level is the weakest claim and the easiest to check. If the composite is GDP-weighted, as the Countries tab states its gauge is, then the four smaller economies carry small weights and their contribution to the level is bounded by those weights. This is arithmetic, not research, and it is usually where the conversation should have ended.
Contribution to timing is the interesting claim. A small input earns its place if it turns earlier than the aggregate often enough to shift the timing of a state change. Smaller open economies face the global cycle before the large domestic-demand economies do, which is a mechanism rather than a finding, and it is the mechanism worth testing.
Conditional contribution is the subtle one. An input can be worthless on average and valuable in the specific states where the big four disagree with each other. When the dominant institutions are aligned, the aggregate is determined regardless. When they split, a tiebreaker with independent information has leverage far beyond its weight. This is the version of the hypothesis most likely to be true and least likely to be tested, because it requires conditioning and therefore shrinks the sample.
The test, and the honest limits on it
Nested models, and you have to build both of them yourself.
The module publishes its input families at the family level, being aggregate central bank balance sheets, global M2, USD liquidity indicators and credit spreads. It does not publish the weighting scheme or the exact series for the headline composite. So you cannot reproduce the vendor's number, and any test you run is a test of the concept on your own reconstruction, not an audit of their series. State that in the first line of the writeup and it stops being a weakness. Fail to state it and the whole document is undermined by the first person who asks how you got the weights.
Build a four-bank aggregate and an eight-bank aggregate from public balance sheet data on the same weighting rule, converted to a single base currency on the same convention. Then evaluate the incremental content of the second against forward returns on the assets your overlay actually trades, at the horizon your overlay actually holds. Not a generic equity index at a generic horizon.
Four design points decide whether the result is worth anything.
- The unit of observation is the regime spell, not the day. Liquidity aggregates are persistent, consecutive daily observations are not independent, and an incremental R-squared computed on daily data will look decisive when the effective sample is a handful of episodes.
- Test the four smaller banks as a block first. Testing them individually gives you four chances to find significance and guarantees you find one.
- Run the conditional version explicitly, restricted to periods where the big four disagree in sign, and report the sample size of that subset alongside the result. It will be small.
- Pre-register the specification before you look at the output, in a dated file. Not for rigour theatre, but because the temptation to adjust the horizon after seeing the first result is overwhelming and undocumentable after the fact.
What to do with each possible answer
Both outcomes have an action, which is the test of whether the research was worth running.
If the four add nothing measurable, do not go and build a private four-bank composite to replace the vendor's. The gain is negligible by construction, since the extra weight was small, and you would trade a maintained product for a maintenance obligation. What changes instead is your language. The smaller institutions stop appearing in attribution paragraphs, the memo template drops them from the driver list, and the desk stops spending meeting time on policy decisions that cannot move the number. That is a real saving in attention, which is the scarce resource on a macro desk.
If one of them does add something, the immediate next question is what it is a proxy for. A small open economy carrying incremental information about global liquidity is rarely doing so because of its own monetary policy. It is more likely standing in for a global factor that your composite does not otherwise capture, such as commodity cycle exposure, dollar funding conditions or safe-haven currency flows. Find the underlying factor and consider tracking that directly, because a proxy tracks its target only until the relationship that made it a proxy changes.
Reporting a null result so it survives
The output most likely to be true here is a null, and nulls are institutionally fragile. They get filed, forgotten and re-litigated in a meeting eighteen months later by somebody who has just read something about the Swiss National Bank.
Write it so that does not happen. One page. The question, the specification as pre-registered, the sample and its effective size in spells, the result including the conditional subset, and an explicit statement of what evidence would reverse the conclusion. That last line is what makes a null durable. "We would revisit if the four smaller banks' stances diverged from the big four for a sustained period, which did not occur in our sample" is a sentence that answers the question when it is asked again, and closes it.
File it with the input inventory rather than in a research folder, so that the next person who reviews composite construction finds the test attached to the thing it was about.