Does Pooling Two Positive-Slope Datasets Guarantee a Positive Slope?

Reported in interviews at GTS

Problem statement

You run a simple (univariate) OLS regression of $y$ on $x$ on two separate datasets, and on each one the estimated slope $\hat{\beta}$ comes out greater than 0. Now you pool the two datasets and run one regression on the combined data.

Is the pooled slope necessarily greater than 0? Explain.

Show hint

Can you settle a 'necessarily' claim faster by trying to prove it or by hunting for a counterexample?

Relevant roles: Data Scientist, Applied Scientist and ML Engineer.

Related free problems