[Bug, sim] When prob(up) < 0.5, trader profit is always zero

[trentmc]

The issue

Via #684, I just created a new plot:

2d scatterplot of (a) model's prob_up vs (b) trader profit.

An example plot is below. It's from running: pdr sim ppss.yaml

The values with prob(up) > 0.5 make sense.

But for prob(up) < 0.5, they're all zero. Something's up.

Example plot

[trentmc]

The problem was:

In a pred_down scenario, it bought back the same # usdcoins, that it had initially sold the tokens for. So its tokcoin balance was improving (if predicting well) but not its usdcoin balance.

This is incorrect, assuming we want to improve the usdcoin balance.

To be correct: It needed to buy the same # tokcoins.

In the PR, the key fix is when the trader exits the trading position in the case of not pred_up.

Here's the old code:

            self._buy(trueprice_eth, usdcoin_amt_spend=usdcoin_amt_recd)

New code:

            # we'd sold, so buy back the same # tokcoins as we sold
            # (do *not* buy back the same # usdcoins! Not the same thing!)
            target_tokcoin_amt_recd = tokcoin_amt_send
            p = self.ppss.trader_ss.fee_percent
            usdcoin_amt_send = target_tokcoin_amt_recd * (1 - p) * trueprice
            tokcoin_amt_recd = self._buy(trueprice, usdcoin_amt_send)