I have access to daily vol quotes for EURUSD options from 2006 to today. I was playing around with them and constructed a "daily rolled backtest" for various options constructs, like straddle, 25 delta risk reversal and 25 delta butterfly. This means I am buying, say a 1M straddle today, hold it for one trading day and my pnl would be the difference between today's and tomorrow s price. I am using the full term structure to price the options one day later, so I am considering the vol roll down. When I do this, the pnl series for straddle and risk reversal are fairly well behaved but for butterfly I get some crazy sharpes of about -5 or less for the 1M and 2M tenors. Shorting the butterfly gives me almost a straight line up in the backtest, which looks very weird. This would mean that FX convexity is heavily overpriced. If I compare this to S&P for example, things look very different. Shorting the butterflies performs rather poorly there. I am aware that this strategy I am looking at here is not tradable since the transaction costs of daily rolling are prohibitive, but can this be true? If I can find a more realistic way of rolling, I could probably still get a very profitable strategy? If convexity in EURUSD is really so heavily overpriced compared to other underlying, what is the reason?