A pure random walk with no drift has an expected next move of zero regardless of what happened before. Charts of random walks show convincing trends, support levels, head-and-shoulders patterns and double bottoms, because human pattern recognition finds structure in noise.
This is the single most useful control experiment in systematic research: run your strategy over simulated random walks with the same volatility as the real series. If it produces similar-looking equity curves and occasional great backtests, your real result may be nothing more than one of those draws.
Markets are not exactly random walks. They are close enough that any claimed edge must be demonstrated against this benchmark rather than against a flat line.
Related: efficient-market-hypothesis, martingale-process, synthetic-data, unit-root