Interesting article! I don't want this to look like I'm trying to poo poo on the parade and I'm 100% in favour of the sentiment about not needing a "PhD in Maths from the Uni of Douchebags" to understand this stuff but the accumulation of the issues below can lead to huge losses. In the example you give of selling Jun20 calls on the SPY I don't even want to imagine how much you would have lost.
1. When discussing the idea of delta hedging and using IV as a distributional measure you should be using *log* returns (not the .pct_change() function in there). This is because in a pure Black-Scholes model the distributional assumption is that log returns are normally distributed. If simple returns were normally distributed then in high vol regimes you could have your model spitting out >-100% price changes and so negative prices.
2. ***This is a hugely risky strategy.*** If i've read it right the idea of the article is to sell short-dated options (with their associated day 1 delta hedge) to collect the volatility premium. In practise this amounts to synthetically creating the breakeven straddle e.g. if you collect 2% for selling your option then as long as the stock doesn't expire >+/-2% away from the stock price you bought it on you should net collect. Many will argue that this is a good idea as the volatility premium tends to persist and many big vol hedge funds do this - they just sit and sell convexity all day. But this is a fundamental misunderstanding of the distribution of variance - sometimes you get periods of low vol (where this strategy works) and other times it will explode leading to you losing everything you have previously made. Which leads to...
3. Options are generally quoted for convenience in IV and ‘cheap’ and ‘expensive’ options are often assessed by looking at implied vs realised. However this masks the issue that options are not actually exposed to volatility, but actually to variance and variance is much more explosive. In other words, if you make 1 100 times in a row and then lose 50 once you still are up 50 - in volatility land. In variance land (square everything) you've just lost 2,400. Again, it's the big moves that count and in this strategy you are selling them.
4. GARCH is fun but not useful for forecasting vol. It's nice autoregressive feature will mean it looks like it decays nicely after a big move (so the forecasted vol for the period following a big move will remain elevated) but it will never forecast you the first move of a huge sell-off - which as mentioned above is critical. In other words, it won't be able to tell you if the big Friday sell off is the beginning of a COVID March 2020 style sell off or just a blip that will be fine by Monday.
5. It's true that longer dated contracts do generally have higher IV, but this isn't always the case. This is especially not the case in the data you are looking at for Jun20 when the term structure was comically inverted.
