The uncomfortable question after a strong year is whether the portfolio earned it or simply held the right exposure while global liquidity expanded. Most attribution decks answer a different question, breaking performance into sector, selection and allocation, all of which are conditional on a market environment the deck never mentions. If the environment did the work, the sector attribution just tells you which part of the environment you happened to be standing in.
The question a committee should be asking
Frame it as a decomposition. Portfolio return over a period is the sum of what any comparable exposure would have earned given the liquidity regime, plus whatever the manager added on top. The first term is liquidity beta. The second is what the fee is for. A regression is the cheapest way to draw the line between them, and it produces a number the committee can carry from one review to the next instead of a narrative that gets rewritten every quarter.
The Global Liquidity Scorecard is a reasonable right-hand-side variable for this because it is defined and it is stable. It aggregates the Fed, ECB, BoJ, PBoC, BoE, SNB, BoC and RBA, plus global M2, USD liquidity indicators and credit spreads, into a composite on a 0 to 100 scale, currently printing 85 with a regime label of RISK-ON and a policy label of EASING. The value of a published composite here is not that its weights are optimal. It is that they do not change based on how your year went, which is exactly the property an attribution variable needs.
Specifying the regression so the answer means something
Four specification choices decide whether the output is informative or decorative.
Use changes, not levels. The composite is bounded between 0 and 100 and spends long stretches in a range, so a levels regression against returns is a spurious-correlation generator. Regress monthly portfolio excess return on the monthly change in the composite. The coefficient then has a clean reading: the return associated with a one point move in global liquidity conditions.
Include the market. A univariate regression on the liquidity change will load up whatever it shares with equity beta, and you will read that as a liquidity effect. Run the two-factor version, portfolio excess return on market excess return and on the change in the composite together, and report the pairwise correlation between the two regressors alongside the result. If that correlation is high, your standard errors are inflated and the split between the two betas is unstable, which is itself the finding and belongs in the memo.
Decide on lags deliberately. A contemporaneous regression is descriptive and answers the attribution question directly, which is what a performance review needs. A specification that lags the composite by one month is a different and much stronger claim, because a significant lagged coefficient implies the signal was tradable ex ante. Run both, label which is which, and never let the descriptive version get quoted as evidence of predictability.
Handle the standard errors. Monthly return series carry heteroskedasticity and, if you use overlapping windows anywhere, serial correlation. Use Newey-West errors with a lag length matched to the overlap. If you have run quarterly returns on a monthly step, the raw t-statistics are inflated by construction and are not reportable.
Reading a coefficient without over-claiming it
The intercept is the interesting number and it is also the fragile one. Annualise it and it is your estimate of return not explained by market beta or liquidity beta over the sample. That is a defensible working definition of the part attributable to process.
Then take the confidence interval seriously. With three years of monthly data you have 36 observations and two regressors. The standard error on an annualised intercept in that setting is routinely wide enough that a point estimate of 300 basis points sits inside an interval spanning zero. The correct sentence for the memo is that the sample cannot distinguish this from zero, not that the strategy generated 300 basis points of alpha. Five years of monthly data, 60 observations, is the point where this exercise starts producing intervals narrow enough to argue about.
Where the sample is too short, fall back to the categorical version. Split the months by the regime label the module was showing at the time, compare mean portfolio returns across the two groups, and report the difference with its standard error. It throws away information and it is far more robust, and for a committee that has to make a decision on 24 months of data it is the honest instrument. A manager whose entire outperformance sits in the RISK-ON months has told you something about the book regardless of what any regression coefficient does.
One framing to reject explicitly. A high liquidity beta is not a criticism. Many mandates are supposed to be long risk and therefore long liquidity conditions, and the beta is the mandate working. The finding that matters is a mismatch between the beta the portfolio actually ran and the beta the mandate described, or a fee schedule priced for alpha on a return stream that decomposes into beta.
The vintage data problem you have to solve first
Here is the constraint that will determine whether you can run this at all, and it needs to be settled before anyone builds a model.
The composite shows what conditions look like now, refreshed at 08:05 AM against inputs that publish weekly and monthly. Trade Signals sits in the module alongside Dashboard, Regime, Risk, Countries, Data and Trade Analysis. In the capture I am working from, that view shows the header, the composite and the two labels. It does not show me a history, an export control, or a series I can point a regression at, so I am not going to assert that a downloadable point-in-time file exists. Confirm what the Data view will actually hand you before you scope this piece of work.
Two distinct problems hide in that. The first is availability, whether any history exists in a form you can consume. The second is vintage, and it is the one that quietly ruins attribution studies. Monetary aggregates and balance sheet figures are revised. A composite recomputed today over restated inputs is not the series anyone saw at the time, and regressing your returns on a restated series builds look-ahead into the explanatory variable. The effect is directionally predictable: revised data is smoother, so it explains realised returns better than the live series did, which mechanically shrinks your estimated alpha and overstates liquidity beta.
The fix is unglamorous and you should start it today rather than when the study is commissioned. Capture the composite, the regime label and the policy label at a fixed time each week into a table with an observation date, and never overwrite a row. Three quarters of that gives you a genuine point-in-time series, and it is the only version of this variable that can support an ex ante claim. Until you have it, restrict yourself to contemporaneous attribution and say so in the footnote.
What it changes in the review, and in the fee conversation
The output of this work is not a chart. It is three sentences that go into the quarterly pack every quarter, in the same form: the portfolio's estimated liquidity beta over the trailing window, the annualised intercept with its interval, and the share of months in each regime label during the period.
Kept consistently, that does two things a normal attribution deck cannot. It makes the environment an explicit line item, so a good year in an expanding liquidity regime is described accurately at the time rather than reinterpreted after the next bad one. And it converts the hardest conversation in manager oversight, the one about whether to cut a manager after two poor quarters, into an evidence question. A manager whose intercept has been stable while liquidity beta dragged the reported number down is in a different position from one whose intercept has been drifting toward zero for six quarters, and only the decomposition distinguishes them.
It also disciplines the fee discussion, which is where this usually ends up. If a strategy's return stream is largely explained by a publicly available liquidity composite and a market factor, then a passive replication of that exposure is the relevant benchmark, and the fee is being charged on the intercept whether or not the term sheet says so. That is a conversation worth having from a regression output rather than from an impression formed in a meeting.