QuantConnect

What a platform like this is

A code-based research platform gives you three things in one place: historical data, a programming interface for writing a strategy, and an engine that steps through the data one bar or tick at a time, placing simulated orders as your code asks. Some platforms of this type also connect to a broker or exchange so the same code can trade live. Whether a particular product does that, and with which brokers, markets and data, is a fact that changes. Read the vendor's current documentation for it, and do not rely on this page or on a review that is a year old.

The appeal is control. You decide every rule, and you can test it before you risk money. The cost is that you must write correct code and understand what the test is actually measuring. A bug in your code, or a quiet assumption in the engine, produces a result that looks like evidence and is not. For the general idea of testing a strategy on past data, see backtesting and the list of backtest software features to compare.

The general risks of this kind of product

A backtest without realistic costs

Every trade pays a spread or commission and usually some slippage. A strategy with a small edge per trade and many trades is the most exposed. If the test assumes fills at the quoted price with no cost, it flatters exactly that kind of strategy. See slippage and spread.

Looking at the future by accident

If your code uses a number that was not known at the time of the trade, such as a daily high used before the day ends, the test cheats without telling you. Platforms try to prevent this, but your own data choices can still leak. See look-ahead bias.

Searching until something fits

Because testing is quick, it is easy to try hundreds of settings. The more you try, the more likely the best one is luck. Keeping part of the data unseen, and testing forward on new data, is the usual defence. See forward testing and walk-forward testing.

Data that is not what it seems

Historical data can have gaps, adjustments for splits and dividends, and lists of instruments that leave out the ones that stopped trading. A test on today's survivors overstates what a strategy would have done. See survivorship bias.

Worked example: what costs do to a backtest

Everything here is made up. A strategy makes 400 trades a year. Before any cost, each trade makes an average of 0.12 percent. The test either ignores costs or charges 0.10 percent per round trip for spread and slippage. To keep the arithmetic plain, the yearly figure is the simple sum of the per-trade percentages, with no compounding.

Calculated by us on 31 Oct 2026, exact. The cost of 40 percent over the year is larger than the whole after-cost result. A small change in the cost assumption moves the result a lot: at 0.12 percent cost the same strategy would show zero. The platform did nothing wrong here. The result simply depends on an input that you choose, and that is why you should always look for where costs are set in a test.

How it hits bots

• Grid and martingale logic can look smooth in a test that stops before a long one-way move. A test that ends early hides the event that ends the bot. See martingale.
• Fast bots depend on fill assumptions most. A test that fills every order at the quote is least reliable for them.
• Live versus tested: when code moves from a test to a live account, order handling, data delays and rejected orders become real. The first live weeks are a test of the plumbing, not only of the idea.

What to verify on the vendor's current pages

Do not rely on this page for any fact about a product. Check, on the vendor's own pages and on the date you read them:

• Which markets and data the platform covers, and how far back the data goes.
• How fees, spread and slippage are modelled by default, and whether you can change them.
• Whether live trading is offered, with which brokers, and what the rules are.
• What the plans cost and what each limits. Prices change, so read the current page and note the date.
• What the terms say about losses, data errors and who owns the code you write.
• Whether results shown in public examples include costs, and over what period.

What this page does not tell you

It does not say whether the platform is good for you, and it states nothing about its prices, features or users because we have not run it. It says what a result from any platform of this kind can and cannot show. The numbers are made up and nothing here is a recommendation.