I'm struggling with some anomalous behavior in an analysis I'm running and was hoping for some advice/insights. I'm attempting to extract the implied funding/borrow costs from ETF option prices (say SPY options) using put-call parity. My method is as follows:
1.) "Europeanize" the options by taking a reasonable parameterization of an implied/local vol model (say SVI), using this model to price Euros and Americans at the same tenor/strike/etc as my market data and subtracting the difference between the modeled American and modeled European from the market American prices.
2) Using these pseudo-european prices, regress put call parity, importantly using the current market expected fixed dividend schedule for the underlying and a multi-stripped rate curve from a number of market rates, swaps, and derivative contracts.
3) From here, it is straightforward to use the regression results to rip out the implied funding/borrowing/some-other-spread-to-the-bank-rate part of the equation. For clarity, I am regressing against strike, leading to
$$X = -e^{(r+\delta) T}[C-P] + e^{(r + \delta)T}[S_0 - \Sigma e^{-r_t t}D_t]$$
and attempting to infer $e^{\delta T}$. The resulting numbers are nonsense for short maturities (+/- 10%, 20%, even 30% I'm seeing), however the long-maturity asymptotic behavior is very much in line with, say, SPX funding costs. Could this be step 1 modeling error in "europeanizing" of short-term options, some operational/microstructure effect of hedging short-dated ETF options, or something else?
