Algo assim deve fazer o trabalho SQL Fiddle
Encontra ilhas de dados sequenciais com o mesmo valor para
SIGN
e aloca-lhes o mesmo valor de agrupamento usando a técnica de número de linha de Itzik Ben Gan, em seguida, agrupa-os e agrega-os. Os CROSS APPLY ... VALUES
desarticula o MIN
e MAX
;WITH T1
AS (SELECT *,
ROW_NUMBER() OVER (PARTITION BY SIGN(PctGain)
ORDER BY WSeqKey) - WSeqKey AS Grp
FROM YourTable),
T2
AS (SELECT MIN(WSeqKey) AS BeginSeq,
MAX(WSeqKey) AS EndSeq,
SIGN(PctGain) AS Sign
FROM T1
GROUP BY Grp,
SIGN(PctGain))
SELECT CASE Sign
WHEN -1 THEN 'Negative'
WHEN 0 THEN 'Equal'
WHEN 1 THEN 'Positive'
END AS [Sign],
Descriptor,
SeqKey
FROM T2
CROSS APPLY (VALUES('Begin', BeginSeq),
('End', EndSeq)) V(Descriptor, SeqKey)
ORDER BY SeqKey