The proper way to show that some rule is wrong is to find and show counter examples, so I did.
The same thing shows regularly in the record, or in any auto-correlated random series. Here are a couple more (using the values you should enter in Wood for Trees) The window is from the start of first year to the start of the second.
1969-1979: trend 0.084
1979-1989: trend 0.063
1969-1989: trend 0.129
1988-1998: trend 0.036
1998-2008: trend 0.102
1988-2008: trend 0.195
Here’s one in reverse.
1964-1974: trend 0.282
1974-1984: trend 0.418
1964-1984: trend 0.160
This isn’t obfuscation. This is an attempt to help show you why your rule is mathematically incorrect.
You can look for an additional condition which would allow your rule to hold. (You spoke of needing no “flip”; but you’d have to define that before I could tell if the condition is mathematically sufficient.) The point is that this “flip” or whatever it is shows up quite a lot in the record.
Cheers — Chris