TRADING 101 / TESTING A STRATEGY
Paper trading vs backtesting: what does each test?
A backtest replays the past. A forward paper test records simulated decisions as new market data arrives. Both can teach you something; neither proves that real trades will perform the same way.
A backtest asks what a rule would have done.
Run a strategy over stored prices or order books, with explicit assumptions about costs and execution. This can expose programming mistakes and help investigate an idea quickly.
The trap is choosing the rule after seeing the answer. If you try enough variations on the same history, some can look good by chance. Using information that was unavailable at the decision time is look-ahead bias. Testing only assets that survived leaves out failures.
A forward paper test asks what it does next.
Fix the rules before evaluation, then record decisions using new observations. Purchases, sales and balances are simulated. This tests whether the pipeline behaves as specified and whether the idea survives a period it was not tuned against.
Changing a rule halfway through creates a different experiment. CouchChange preserves the earlier version and its results instead of rewriting its history.
Example: a rising price can still lose money.
Suppose a hypothetical $100 entry budget includes its buying fee. Under our teaching calculator’s starting assumptions—0.8% fees and a 0.05% adverse-price allowance on each side—a 1% price rise returns about $99.30 after selling. The modeled loss is about $0.70.
That is arithmetic under specified assumptions, not a venue quote or a recorded trade. The example omits spread, market depth, network fees, delays, failed orders, taxes and hosting.
A simulated fill is still an assumption.
A displayed price does not prove an order could complete at the intended size. Real orders can arrive late, fill partly, move the market, or fail. A stop condition requests an exit; it cannot guarantee the price received.
Our spot study waits for a later order-book observation before modeling a fill, with explicit fees and execution allowances. This adds discipline, but it does not reproduce every consequence of a real order.
Read a result with five questions.
- Was the rule fixed before these outcomes were observed?
- Are costs, open losses, missing data and incomplete exits included?
- What happened to a simple holding or cash comparison?
- Are the observations independent, or repeated views of the same move?
- Does the result hold up on a later period and under worse execution assumptions?
A win rate or a good-looking chart cannot answer those questions by itself. New evidence can weaken an idea that once looked promising.
CouchChange runs unfunded forward paper experiments. Public study pages describe the designs. Recorded market results remain owner-only while publication rights are reviewed. No simulated balance can be withdrawn.
Inspect a test you can follow.
Open the BTC, ETH and SOL trend study to see the decision and fill process, or read our research method. For the broader vocabulary, return to Trading 101.