With an odd (3-year) period, each moving average centres directly on the middle year of its group of three, but the first and last years of the series are left without a trend value.

Step 1 — form 3-yearly moving totals (each total covers a year, the year before it, and the year after it):

Step 2 — independent check. Re-adding each triplet directly from the original data confirms the totals: $20+22+27=69$, $22+27+25=74$, $27+25+31=83$, $25+31+29=85$, $31+29+36=96$ — all match, so the trend values above are confirmed correct.

Note that no trend value can be computed for 2016 (no preceding year available) or 2022 (no following year available) — this is an inherent limitation of the moving average method, not a computational gap.

Trying to force a trend value for the first and last year of the series by averaging over fewer than 3 years — this is not the 3-yearly moving average method and should simply be left blank, honestly.