System study · 2y daily returns

System Map

Network analysis identifies the system; the regime model reads its state.
Members: UNP, WAB, GBX, TRN, CSX, NSC

Step 1 · Who moves together

daily return correlation
UNPWABGBXTRNCSXNSC
UNP1.000.490.350.410.720.82
WAB0.491.000.500.500.460.47
GBX0.350.501.000.630.360.36
TRN0.410.500.631.000.400.40
CSX0.720.460.360.401.000.77
NSC0.820.470.360.400.771.00
High correlation means these names are one bet, not several. That is what makes the system thesis true and what makes owning all of them concentration rather than diversification.

Step 2 · Who moves first

|corr| > 0.088 is the ~5% significance line
same day
corr
partner→UNP
best lag
UNP→partner
best lag
read
NSC0.82-0.099lag 4-0.080lag 1NSC leads by 4d
CSX0.72+0.081lag 5-0.106lag 1UNP leads by 1d
GBX0.35-0.074lag 4+0.030lag 5no lead-lag
WAB0.49+0.069lag 5-0.074lag 2no lead-lag
TRN0.41+0.058lag 5+0.053lag 5no lead-lag
A partner that leads UNP is a signal. Daily lead-lag correlations are usually tiny and often noise — treat anything near the threshold as a hypothesis to re-test on fresh data, not a finding.

Step 3 · What state the system is in

equal-weighted basket of all 6 names
System state today
CALM
89% probability
Calm regime
39days
mean +0.07%/day · vol 1.11
Stressed regime
2days
mean +0.62%/day · vol 3.88
System basket
+53%
501 days · 22% ann vol
Fitted 2-regime Markov switching model (statsmodels). The regimes were inferred from the data — nothing was hand-labeled. Expected durations come from the estimated transition matrix, same 1/(1-p) idea as your rule-based dashboard.