Let's now suppose that we bought one share of every IPO at the exact opening tick and sold them at the precise closing price listed in these figures; what would our returns look like in terms of dollars earned?

To answer this question, let's look at the actual dollar price change from open to close:

df['1st Day Open to Close $ Chg'] = (df['1st Day Close'] - df['Opening Price']) 
 
df[df['Year']==2018].sum() 

The preceding code generates the following output:

From this, we see the first day open-to-close dollar total is just above $28. That number is for over 173 IPOs so far in 2018:

df[df['Year']==2018]['1st Day Open to Close $ Chg'].describe() 

The preceding code generates the following output:

That reflects a mean first-day gain of just over 16 cents per IPO. And remember, this is under ideal conditions where we ignore transaction costs and slippage.

Let's now take a look at what the distribution of returns for these IPOs look like. This may help us understand how to improve our returns over the baseline Naïve Bayes strategy of just buying every IPO:

fig, ax = plt.subplots(figsize=(16,8)) 
df[df['Year']==2018]['1st Day Open to Close % Chg'].plot(kind='hist', bins=25, ax=ax) 

The preceding code generates the following output:

We see that returns are centered around zero, but there is a long tail to the right where there are some exceptional returns. It would be quite financially rewarding if we were able to identify some of the commonalities that these exceptional IPOs have that we could capitalize on.

Let's see whether we can use machine learning to help improve our results from a Naïve Bayes approach. A reasonable strategy would seem to be targeting that long tail on the right, so we'll focus on feature engineering in the next section.