PairScanner

Case study · US bond ETFs

MBB
/ SPMB

Two ETFs, one benchmark. This example starts with an identifiable economic relationship, then examines it using PairScanner’s statistical tests.

Historical snapshot as of
Analysis: PairScanner

Correlation
0.98
Daily returns
P-value
0.0082
Test threshold: 0.05
Closing Z-score
-0.15
Window: 60 sessions
Coefficient β
0.9950
Log-price regression

Period studied : 2021-07-27 → 2026-09-08 · 1284 shared sessions

Why these two ETFs are related

MBB (iShares MBS ETF) and SPMB (State Street SPDR Portfolio Mortgage Backed Bond ETF) track the Bloomberg U.S. MBS Index, an index of U.S. agency mortgage-backed securities. This shared benchmark motivates the comparison; it does not, by itself, prove cointegration.

Compare the paths from the same starting point

Both series start at 100 on the first shared date. A level of 120 represents a 20% increase from that starting point, calculated from closing prices adjusted for dividends and splits. The chart compares changes, not dollar prices.

Price paths normalized to 100
MBB · iSharesSPMB · State Street SPDR
80.090.0100.0110.0Price paths normalized to 100Shared dates and starting point. Both curves use the full period shown.
2021-07-272024-02-142026-09-08
Shared dates and starting point. Both curves use the full period shown.

Daily return correlation is 0.98. It describes how day-to-day changes have moved together. It does not establish whether the gap between the two assets returns to a lasting equilibrium.

Test the stability of the relationship

PairScanner estimates β by regression, then constructs the spread S = ln(MBB) − β × ln(SPMB). The Engle-Granger test uses log prices. Its null hypothesis is no cointegration; a p-value below 0.05 rejects that hypothesis at the selected level.

Spread centered on its historical mean
Mean
-0.0050-0.00250.00000.00250.00500.0075Spread centered on its historical meanThe full-period mean is subtracted to place the reference at zero. This retrospective view is not a trading simulation.
2021-07-272024-02-142026-09-08
The full-period mean is subtracted to place the reference at zero. This retrospective view is not a trading simulation.

Here, p = 0.0082. The threshold is met; this ensures neither persistence of the relationship nor a successful trade.

How to read the hedge coefficient

β is 0.9950: it weights SPMB’s log price in the spread. It is not directly a number of shares; translating it into quantities also requires prices and the capital allocated.

Locate the gap at the last close

The Z-score measures the spread’s distance from its rolling mean in standard deviations over 60 sessions. Dashed lines at ±2 are reference levels. Z can still be calculated when the cointegration test is inconclusive.

Spread Z-score and reference levels
Reference levels ±2
-5.0-2.00.02.05.0Spread Z-score and reference levelsThe window includes the observed session. Initial sessions without a complete window are not plotted.
2021-07-272024-02-142026-09-08
The window includes the observed session. Initial sessions without a complete window are not plotted.

At the study date, Z = -0.15, inside the ±2 reference levels. It describes the gap at that close; it does not validate the relationship on its own.

The relationship passes the checks, but the ±2 deviation threshold is not reached at this close: a validated relationship does not mean an immediate entry.

Scope and limits of this reading

Data
Adjusted closing prices used by PairScanner. Shared dates only; missing, non-finite and non-positive observations are excluded. Last observation: 2026-09-08.
Estimation
β and the cointegration test use all 1284 observations in this study. Charts are descriptive and retrospective: β was not known in advance for each session shown.
Out of sample
Software criteria met. A separate β is estimated only on the first 70% (2021-07-27 → 2025-02-21), then frozen. The ADF test uses the spread over the following period (2025-02-24 → 2026-09-08): p = 2.13e-08, against a threshold of 0.05. Half-life is 1.82 sessions, against a limit of 60. It is estimated over the entire history with this frozen β, not just the test period. This single chronological check is not a profitability backtest.
Multiple testing
The displayed p-value belongs to this individual test. It is not a q-value adjusted across all scanner pairs. A p-value is not the probability of a profitable trade.
Case selection
This pair was selected after reviewing historical results to illustrate a positive case, without changing thresholds or the period. This retrospective choice is not evidence of a strategy’s out-of-sample performance.
Execution
Closely tracking ETFs can leave deviations too small to cover bid-ask spreads, commissions, financing and borrowing the shorted asset. Costs, liquidity and distributions are not simulated here. A statistical relationship alone therefore does not make a trade attractive.

What this example helps you understand

Correlation describes daily changes, cointegration tests a relationship between price levels, and the Z-score locates a deviation. These three readings complement each other. This is a dated study: the application’s figures can change after subsequent closes.

This educational study is independent of the pair of the day, which is selected for its trading relevance.

Another result, the same method

Coca-Cola / PepsiCo shows the other case: two familiar names and correlated returns, without cointegration established over the period studied.

Read the Coca-Cola / PepsiCo study →
Also explore: KMI / WMB, an entry-zone Z demonstration →

Explore the scanner’s pairs

Find updated analyses and results in PairScanner.

Open PairScanner

Historical results do not determine future outcomes. This study is not an investment recommendation.