AMZN finally crossed $1000 a share today, an occasion not missed by many media outlets to question whether it is time for AMZN to split. Matt Levine’s take this morning was as follows:

The basic rational discussion about stock splits goes like this:

Our stock is at $200. Is that a problem?

No that is a meaningless number.

Okay should we split it in two so that each share is worth a more manageable $100?

No come on that is dumb.

I mean, it’s fine! It’s fine. Do the split, or don’t, whatever. The price of one share of stock doesn’t matter, and it is silly to spend much time thinking about optimizing it.

I agree that share price is meaningless for most purposes, but price does matter when it comes to orderly trading practices. On Friday on CNBC, Eric Chemi recounted a conversation he had with me earlier in the year about how nosebleed share prices can increase the chances a stock will be manipulated.

To explain why, consider this side by side comparison of trading in AMZN and MSFT, two stocks in the same sector with similar market caps, at exactly 11:00 am Friday:

The main difference, of course, is that AMZN trades in the $1000 range while MSFT trades in the $70 range. If share price really doesn’t matter, one would expect most measures of trading performance to mimic the roughly 14:1 ratio of share prices. But that isn’t what we see here. Note the following:

1. The bid-ask spread in MSFT is always one penny. The spread in AMZN is 34 cents at the moment shown but had been 66 cents a few seconds earlier (you can pull up these charts yourself and move around in them for free here). The spread in AMZN is considerably worse than the 14:1 ratio you would expect if price levels were meaningless. Wide spreads mean higher costs for investors.

2. The order book in MSFT shows several thousand shares quoted at every penny price tier on both sides of the market. The order book in AMZN shows wide gaps of eligible price tiers with no shares quoted at all (e.g., the second best bid is 33 cents lower than the best bid), and where there are quotes they are often odd lots (less than 100 shares) which are not protected from trade throughs under Reg NMS. To buy $2 million worth of MSFT (about 28,000 shares at these prices), assuming you could access what was quoted, you could expect to have to pay 3-4 cents through the market in these conditions. To buy $2 million worth of AMZN (2,000 shares), you would have to go 19 tiers into the book and pay prices over $1 away from the best bid. Again, much worse than the 14x multiple one would expect if price didn’t matter.

3. The thin order book in AMZN makes it more vulnerable to market manipulation. Spoofing and layering are manipulation schemes in which a trader deceives other traders as to the true price of a stock by displaying artificial bids or offers that disrupt the natural balance of supply and demand. In listed stocks and futures, the balance of supply and demand can be seen visually in the shares quoted on each side of the order book. A spoofer puts his thumb on that scale, causing prices to tilt in his favor. In MSFT, we see around 75,000 shares quoted in the top 10 tiers of each side of the order book. To add a 20% thumb on the scale here, a spoofer would have to enter orders for around 15,000 shares, which would expose him to over $1 million of risk. In AMZN however, where only 1,000 shares are quoted in the top 10 tiers, a 20% thumb on the scale would require only 200 shares, or $200,000 of risk. It is thus 5x cheaper to spoof AMZN than it is to spoof MSFT.

So what should AMZN do? To have a robust well-functioning order book, with tight spreads, deep liquidity, and inhospitable conditions for spoofing, large cap companies like AMZN are better off with share prices under $100. Share prices under $30 lead to a different issue with artificially wide spreads caused by the one penny minimum tick size in Reg NMS, so the optimal range is $30-$100. AMZN and the other large cap stocks priced above $100 would do their investors a favor by annually splitting to keep share prices in that range.