Sobel's test (aka, testing with delta-method standard errors) is known to be underpowered. Indirect effects calculated as the mean of two coefficients have an asymmetric sampling distribution, so any method that gives symmetric confidence intervals (which includes any method based on standard errors) will give a flawed answer. There are several approaches that generate more accurate, asymmetric confidence intervals. David MacKinnon and colleagues have done much work on this over the last 10-15 years.
All that said, the Sobel formula is very simple and easily implemented in spreadsheet software if SPSS doesn't have a function.