Methodology
This widget summarizes the series using three simple signals computed from its own history: Direction (short-term slope of the level), Core acceleration (recent pace vs structural pace), and Trend shift (recent change in the YoY trend).
1) Convert the level into monthly change
For level-type series (index/price/amount), we compute the monthly log change (≈ % m/m):
g(t) = 100 × [log(x_t) − log(x_{t−1})]
2) Direction (slope of log-level over 6 months)
We estimate the slope of the log-level over the last 6 months using a linear fit:
y(t) = 100 × log(x_t)
y(t) ≈ a + b·t
Direction = b (≈ % per month)
3) Core acceleration (recent pace vs structural pace)
We compare the average monthly pace over the last 12 months to the last 36 months:
mom12 = mean(g, last 12 months)
mom36 = mean(g, last 36 months)
accel = mom12 − mom36
4) Trend shift (YoY trend change over 3 months)
We compute the YoY rate, smooth it with a short EMA, and measure how much it moved over the last 3 months:
YoY(t) = 100 × (x_t / x_{t−12} − 1)
Trend(t) = EMA(YoY, 3)
Δ3m = Trend_now − Trend_3m_ago
Scaling (auto range)
Gauge limits are set from recent history using a robust percentile of |signal|.