Further to my point, testing can only "prove" an edge is no longer effective ... during the test period. Not that it can no longer be effective at all ever again.
You see some algo's and you know they're killin' it. I see it in the rate flys and all the DV01 spreads in futures, and the index spreads where the discount offers a capital efficient hedge, e.g. cutting margin on a large position with relatively few NQ contract. DJX/RUT comes to mind on a weekly. They got some real nice algo's guys.
It's too late, I am reforming myself at this very moment. I will begin preying on innocent beginners from now on. False breakout fading is The Holy Grail!
Edges are structural. When the structure changes, the edge changes. Markets are used to monetize edges. Look up SOES bandits as an edge example. A interesting story.
Alpha = something that pays you with excess returns Edge = your workflow that gets you that alpha Simplistic example: There doesn't seem to be much alpha in distressed stocks (Is there alpha in distressed stocks? (U.S. evidence) (alphagenstrategies.com)). However, you have a strategy where if you filter for size and industry, you historically generate very positive absolute returns (e.g. you go long distressed above a certain size and excluding certain sectors, short the rest). This defines your strategy to generate alpha (excess returns). The filtration process is an analytical edge you have over the market. In that simplistic example, assuming you are using standard data, then your edge in the filtration process won't last for a long time (everyone else will find it and start to bid up those stocks too), so over time your edge decays and your alpha no longer pays.
The only enduring edge is taking candy from babies. If the market ever runs out of them, we're all done here!
The problem is that these babies are not the retail traders like you and me, but the Wall Street. They can never stop whining like little bitches. Always whining and crying. Then turn around and talk as if they're the smartest in town.