Deleting rows which sum to zero in 1 column (like pairs) but some elements in other columns are different

We aggregate by city and by abs(value) to match transactions, then add them up and remove those matched transactions that add up to 0:

df2 = (df.groupby([df['city'], abs(df['value'])])
         .agg({'product':'first','city':'first', 'date':'first','value':sum, 'reference':'first'})
         .reset_index(drop = True)
)
df2.where(df2['value']!=0).dropna()

output:

    product city    date        value   reference
0   t       Munich  05-01-2020  2.0     jidi1
2   z       Munich  21-01-2020  30.0    212jio
3   y       Paris   21-01-2020  -16.0   xji