Use adjusted close for a simple backtest based on returns. For a realistic trading simulation, use actual prices and account for corporate actions explicitly.

The regular close is the price quoted that day. Adjusted close accounts for splits and distributions so prices can be compared over time. Those historical values are synthetic and useful for calculating returns.

Consider two examples:

- A stock closes at $100 before a 2-for-1 split and $50 afterward. You now own twice as many shares. Comparing raw closes would incorrectly show a 50% loss.
- A stock closes at $100 before going ex-dividend and $99 afterward, with a $1 cash dividend. An eligible shareholder has $99 of stock plus a $1 dividend entitlement. The price decline alone does not capture their total return.

Choose the data to match what you are testing:

| Backtest purpose | Appropriate data |
|---|---|
| Buy-and-hold performance or a strategy based on percentage returns | Split-and-dividend-adjusted close, after checking the provider’s methodology |
| Indicators that should ignore mechanical price changes from corporate actions | Consistently adjusted prices |
| Order fills, share quantities, cash balances, and transaction costs | Actual prices, with splits and dividends handled explicitly |
| Rules involving absolute prices, such as "buy stocks below $10" | Prices available at that historical date |

QuantConnect, for example, has raw, split-adjusted, dividend-adjusted, and total-return modes. [Data normalization documentation](https://www.quantconnect.com/docs/v2/writing-algorithms/securities/asset-classes/us-equity/requesting-data)

A few details can affect the results:

- If your return series already incorporates dividends, adding dividend cash again counts them twice and overstates performance. [Corporate-action documentation](https://www.quantconnect.com/docs/v2/writing-algorithms/securities/asset-classes/us-equity/corporate-actions)
- Keep price fields consistent. Comparing an adjusted close with an unadjusted open, high, or low can produce misleading signals.
- For absolute-price rules, use point-in-time data. Today’s adjusted history can incorporate later corporate actions and change whether an old price appears below a threshold. [Look-ahead bias documentation](https://github.com/QuantConnect/Documentation/blob/master/03%20Writing%20Algorithms/01%20Key%20Concepts/10%20Research%20Guide/07%20Look-ahead%20Bias.html)