Three spread estimators from daily bars average a 0.125 to 0.265 time-series correlation with the quoted spread
The Corwin-Schultz estimator builds a bid-ask spread from daily high and low prices. Across 22 currency pairs, its average time-series correlation with the quoted spread is 0.158, 0.125 and 0.265 for the three variants tested, CS1, CS2 and CS3. Squared, that is 1.6 to 7 percent of the variation (my arithmetic).
Li, Lambe and Adegbite measured the correlations (International Review of Financial Analysis, 2018): DataStream pairs, 1990 to 2016, against a monthly benchmark from daily closing quoted spreads. I transcribed Table 15 of the 2017 working paper.
Across pairs at one date the same table gives 0.636, 0.297 and 0.948, so CS3 ranks pairs by cost almost in order and CS2 does not. Through time within a pair, averaged over the 22 pairs, the three top out at 0.265. The paper's reason is that the spread is small beside the mid-price volatility.
I would use a daily-bar estimate only to compare pairs on one date, and read one pair's change in cost from the bid and ask in the tick data.
The averages are over 22 pairs and the paper gives no pair-by-pair figures. The benchmark is a vendor's closing quoted spread and the sample ends in 2016. I have not run the estimator on any broker's feed.


