# Qlib's US Region Does Not Make Commission Per Share

An [open Qlib question](https://github.com/microsoft/qlib/issues/1801) asks how to model US trading costs when the commission is charged per share. The example already sets `trade_unit` to one and passes a small number as `open_cost`. Those settings look close to a per-share fee, but they describe two different things.

The useful first check is the unit of each input. `trade_unit` controls the permitted order increment. `open_cost` and `close_cost` are rates applied to trade value. Changing the region can set US defaults, but it cannot change that fee equation.

## Trace the actual fee path

The [Qlib region configuration](https://github.com/microsoft/qlib/blob/main/qlib/config.py) sets the US trade unit to one and removes the China A-share price-limit threshold. So initialize the provider with the US region:

```python
from qlib.constant import REG_US
qlib.init(provider_uri="path/to/us_data", region=REG_US)
```

Then inspect the [Exchange implementation](https://github.com/microsoft/qlib/blob/main/qlib/backtest/exchange.py). The relevant path is short:

```python
trade_val = order.deal_amount * trade_price
trade_cost = max(trade_val * cost_ratio, self.min_cost)
```

For a buy, `cost_ratio` includes `open_cost`; for a sell it includes `close_cost`. The fee therefore changes with the trade price when the share count stays fixed. A true per-share commission would stay the same in that comparison, until a minimum fee takes over. Setting `trade_unit=1` does not alter this calculation.

## A four-case fixture before changing code

This is the test matrix I would freeze for a custom fee rule. It specifies relationships, not a claimed result from a run.

| Case | Input held fixed | Change | Expected distinction |
| --- | --- | --- | --- |
| Same shares | Share count | Use two different trade prices | A per-share fee stays equal; a value-rate fee changes. |
| Same trade value | Total notional | Use different share counts | A value-rate fee stays equal; a per-share fee changes. |
| Minimum floor | Share count and price | Choose an order below the declared minimum | The configured minimum wins under either model. |
| Cash boundary | Available cash and target order | Put cash just inside and outside the fee-inclusive limit | Fill quantity and charged fee follow one consistent rule. |

The last case matters because the Exchange uses the fee when it clips a buy to available cash. Replacing only the final reported `trade_cost` would make the report look right while the simulated fill quantity still follows the old rate-based calculation. A custom exchange needs one fee policy for both cash admission and the resulting trade record.

## What this check proves

It answers the configuration question in the issue: `region=REG_US` selects US defaults, while the built-in `open_cost` and `close_cost` remain percentages of trade value with a minimum. The four cases above are proposed acceptance tests for a per-share extension. I reviewed the current source path; I have not run a custom Exchange or measured a strategy result here. Broker-specific schedules, taxes, rebates and partial fills need their own explicit rules before a backtest can claim to represent them.