La siguiente solución supone que siempre hay 3 filas por trimestre (solo el último trimestre puede ser parcial), SELECCIÓN única, sin recursividad :-)
WITH sumQuart AS
(
SELECT *,
CASE
WHEN ROW_NUMBER() -- for the 1st month in a quarter
OVER (PARTITION BY CCP, Years, Quart
ORDER BY months) = 1
-- return the sum of all GTS of this quarter
THEN SUM(GTS) OVER (PARTITION BY CCP, Years, Quart)
ELSE NULL -- other months
END AS sumGTS
FROM gts
)
,cte AS
(
SELECT
sq.*,
COALESCE(b.Baseline, -- 1st quarter
-- product of all previous quarters
CASE
WHEN MIN(ABS(sumGTS)) -- any zeros?
OVER (PARTITION BY sq.CCP ORDER BY sq.Years, sq.Quart, sq.Months
ROWS BETWEEN UNBOUNDED PRECEDING AND 3 PRECEDING) = 0
THEN 0
ELSE -- product
EXP(SUM(LOG(NULLIF(ABS(COALESCE(b.Baseline,1) * sumGTS),0)))
OVER (PARTITION BY sq.CCP ORDER BY sq.Years, sq.Quart, sq.Months
ROWS BETWEEN UNBOUNDED PRECEDING AND 3 PRECEDING)) -- product
-- odd number of negative values -> negative result
* CASE WHEN COUNT(CASE WHEN sumGTS < 0 THEN 1 END)
OVER (PARTITION BY sq.CCP ORDER BY sq.Years, sq.Quart, sq.Months
ROWS BETWEEN UNBOUNDED PRECEDING AND 3 PRECEDING) % 2 = 0 THEN 1 ELSE -1 END
END) AS newBaseline
FROM sumQuart AS sq
LEFT JOIN BASELINE AS b
ON B.CCP = sq.CCP
AND b.Quart = sq.Quart
AND b.Years = sq.Years
)
SELECT
CCP, months, Quart, Years, GTS,
round(newBaseline * GTS,2),
round(newBaseline,2)
FROM cte
Consulte Fiddle
EDITAR:Lógica agregada para manejar valores <=0 Fiddle