If you assume price = constant + superimposed sinusoidal curve, with the amplitude of the curve big enough to trigger buying and selling, then by your method, you repeatedly buy low, sell high, over and over again. Which means you make money, assuming the spread and transaction costs are less than what you earn from buying low and selling high.
Any curve can be represented as a sum of sinusoidal curves, i.e. a Fourier series. Therefore, it becomes mathematically provable that your method, if properly implemented, will cause you to benefit from volatility.
I'd like to see that proof

A couple of problems I see with it:
1) you are talking about piecewise sinusoids (right? reset at each sudden price change?). That complicates any kind of frequency domain analysis. Lots of noise.
2) After a price change, how do you determine what phase (and amplitude) to start the next piece at?
If you really did mean fourier analysis of the whole price data, then you would see low frequency cycles with a bit of luck (but too many people already found those, so they're tiny). The sudden price moves add way too much noise to be able to detect anything sinusoidal at day trader frequencies.
I think that I understand that your criticism may encapsulate that humans are way too inconsistent in order to make such a system work mathematically as profitable - however, couldn't you program a bot to take out some of the human error and instead of having it set at really close intervals (like they probably do in china with no fees), they set them at intervals like $10 - or maybe more accurately to use percentage moves, like a .5 or 1% move in one direction triggers a sell, and then every equal increment. Then buy backs would be 1% or more below the sales price. Of course, there are variations about what increments to use and what quantities.
Edit: I wrote the above post before reading BTCtrader71's response. His response does not seem inconsistent with mine, but seems a bit more eloquent and mathematically oriented than mine. hahahaha