SPY skew minus SPX skew Granger causes VIX. Skew Metric is calculated from snapshot data 15min pre-close, so there is considerable measurement nose. With both series at least somewhat autocorrelated, that can lead to some spurious alpha. My estimate is that noise accounts for about half of the 16% r-squared. Still leaves a lot of alpha; as alpha, corr, Sharpe and fraction same sign are all non-linear in R2.
I think the article is reading too deeply into what is mostly a flows-driven story. Part of the the story is the level of vols, people are hedging closer to the money and people are substituting calls for stock. Part of the story is driven by the structured notes (specifically autocallables - as we are rallying, they are covering the short vega hedges). Part of the story is due to suppression of volatility (in local vol setting, one can think of skew as the "gamma of vol" and with vol being suppressed, demand for that convexity is low too). As a side note, the skew realizes much better in RUT/IMW or NDX/QQQ, so the trade to do is to sell SPX skew and buy IWM/QQQ skew.