Common mistakes and myths

Frequency of being right and size of outcomes are two separate numbers. Suppose, illustratively, that a plan is right 9 times in 10, gaining 1 unit each time and losing 20 units on the tenth. Per ten tries that is 9 gained and 20 lost: a net loss despite being right 90 percent of the time. Looking at hit rate alone hides this; the payoff of each outcome matters as much as how often it occurs.

Leveraged positions often come with a margin requirement: if losses reduce the cushion below a threshold, the position can be reduced or closed automatically, sometimes at an unfavourable price and at the worst moment. This turns a temporary fall into a permanent loss, because there is no time left to wait. Some products can lose more than the amount initially put in.

Costs are paid whether the outcome is good or bad, so they act as a steady drag that must be overcome before anything is gained. They also reduce the capital that later positions work with, and a percentage cost applied repeatedly compounds like a negative growth rate. Spread and slippage widen in fast or thin markets, so real costs are often higher exactly when activity is highest.

Flexible rules with many adjustable settings can fit any history. The standard defence is out-of-sample testing: tune on one period, then test once on data that was kept aside and never used for tuning. Walk-forward testing repeats this over rolling periods. The multiple-testing problem is that if many variations are tried, some will look good by chance alone, so the more things tried, the less any single good-looking result means. Statisticians adjust for this, and the honest summary is that a historical result is evidence of what happened, not a measure of what will happen.

Confirmation bias: favouring evidence that supports a prior belief. Loss aversion: losses feeling larger than equal gains. Overconfidence: overrating the accuracy of one's own judgement. Sunk-cost fallacy: continuing because of what is already spent. Gambler's fallacy: thinking a run of outcomes makes the opposite more likely when the events are independent. Recency bias: giving too much weight to recent events. Naming a bias does not remove it, but written rules decided in advance reduce its room to act.

Scams rely on well-known persuasion levers: urgency (act now), authority (an impressive title or screenshot), social proof (others are in), and secrecy (do not tell anyone). Fabricated screenshots and invented track records are cheap to produce, so a displayed result proves nothing. Consumer-protection bodies generally advise checking who is responsible for a service through official channels, never paying into a private account, and taking time before any transfer. If something has already gone wrong, the usual guidance is to stop sending money and report it to the relevant local authority.