You may use part-worth coding, and if you have 5 levels of price (in your conditional pricing table), then that will use up 5-1 parameters to estimate the main effect of price. If you use a linear price, then you will use up just one parameters to estimate for the main effect of price.
Sometimes researchers have enough data and few enough SKUs (brands) in their studies to estimate the interaction between brand and price. In that case, using linear pricing saves a lot of parameters to estimate, because in the case of 12 SKUs and 5 price points in the conditional pricing table (5x12 cells in the table), estimating interaction effects under part worth specification would lead to (12-1)+(5-1) for the main effects plus (12-1)(5-1) for the interactions or 59 parameters to estimate, whereas with linear coding of price it would require (12-1)+(5-1) for the main effects plus (12-1)(1) or 26 parameters to estimate.
If doing linear estimation and conditional pricing, your % change up and down from average price should have followed the same pattern across all SKUs. Let's imagine you have 5 price levels, with relative prices of -20%, -10%, AvgPrice, +10%, and +20%. In CBC/HB software, you select "Linear" for the Coding type for the Price attribute, and then at the right in the "Value" column, you would type the following 0.8, 0.9, 1.0, 1.1, 1.2 to indicate the relative prices used. During CBC/HB estimation, the software will then give you a message that all values for linear terms have been zero-centered.
Then, when you use SMRT simulator or the Online Simulator, you make it much easier on yourself if you check the option that says to use the original conditional pricing table values in the simulator. That way, you can simply specify the original price values from the original table (as was seen by respondents), and the software will map those values according to the table to the relative prices (on the 0.8 to 1.2 scale) prior to multiplying by the utility coefficient. Much more intuitive way to specify prices for products rather than having to convert the prices to 0.8 to 1.2 relative scaling!
Either way, using part-worth estimation (which burns up more parameters to estimate, but has the potential benefit of capturing non-linearity) or using linear price will allow you to interpolate between prices included in the experiment.