I've been thinking a lot about intelligent ways to reduce high frequency trading. Estimates place high frequency trading at 60% of the market's volume, and the problem is that these high-frequency traders are functionally exploiting the insensitivity of long term buyers to penny-sized changes in price to extort away that money. For more details on why high frequency trading is worthless, I've written about it here: http://tfideas.blogspot.com/2009/11/problem-with-high-frequency-trading.html
The problem is with structure. A percentage transactions tax doesn't make a lot of sense, because it punishes long term investors and disincentives investment (which we need more of, not less), and it'll be hard to make it small enough not to hurt all of us but big enough to discourage high frequency trading. It also has the problem where people who earn income and save it will presumably trade with each paycheck, which means you're functionally raising income taxes - even if they're not trading at ALL, they're being taxed. You could deal with this by only taxing the sale of a stock, but sophisticated investors may be able to derivative or option their way out of it and it also harms people like retirees who need their savings to live on.
A flat per-transaction fee doesn't make sense for the same reasons - it discourages smaller investors, and bigger investors are putting enough money in that any flat fee small enough wouldn't work.
So how about this - tax turnover. Everyone is obligated to pay a tax:
T% * (the lesser of their total sales and total purchases minus X% * the highest end-day value of the account observed over the year, net of leverage.)
where T is the tax rate and X is turnover, which is the implicit inverse of holding period when you strip out deposits and withdrawals. For example, a holding period of 2 years would have X=50%.
The reason you use highest end-day value is because you don't want to unfairly reward people who deposit or withdraw money, earn money in the market, lose money or take volatile vs stable trajectories. If you use starting value, you punish net depositers and successful long term investors who turn over once every few years but compound at a fast enough rate that turnover at the end of the year if it's grown since the beginning of the year would get taxed. Similarly, if you use ending value, you punish net withdrawers and people who lost money in the market. If you punish average value, you punish people whose stocks travelled in a U shape, and if you punish the average of start and end prices, you punish people whose stocks travelled in an upside down U. Using the highest observed value is agnostic between depositors, withdrawers, successful investors, unfortunate investors and volatile investors.
The reason the highest-observed value works is because there is a difference between the severity of type 1 and type 2 error here. Generally, the people who you'd like to classify as fast traders are going to have a tax base of well in excess of the X% * highest end-day value threshold. The few who would slip through the cracks will often be the slowest of the fast traders, who presumably have the least impact, and it's not a big deal to not tax them. However, you don't want to end up taxing innocent investors by accident, because that discourages capital markets and real liquidity (as opposed to the fake liquidity from HFT - see the link above). There is a volatility problem - a high frequency portfolio that grows fast, shrinks fast or is very, very volatile may set a high-water mark that is well in excess of their "true" account value for most of the year, but it'd have to be incredibly volatile to avoid paying any tax at all, and probably won't last long. A high frequency strategy that can get deposits or compount at a rate in excess of the speed at which it turns over has never been seen, to my knowledge. If it is, we'll deal with it when we get there.
While I hate the idea of the government deciding appropriate turnover, it already does with the LT vs ST cap gains tax (implicitly, 100%). Because short-term gains and high frequency trading aren't equivalent, you'd probably want a much higher threshold. 400% would imply that you trade on every quarterly earnings, which seems like a reasonable cutoff for anyone to be considered an investor vs. a speculator (I'd argue that even that level is a speculator, but it is conceivable you could be investing for quarterly results, so 400% probably is right). Many HFT strategies can do that in a day (and I believe some can do it in a matter of minutes or hours); almost anything you'd consider high frequency could plow through that in a few weeks.
Even if a HFT strategy starts at $100, "only" turns over 100% a week, and it doubles in a year to hit a high water mark, with 50 weeks a year of trading (9 closed days + i believe 2 half days)... that means it has an exemption of 800 dollars, which it would manage to surpass its threshold (very) roughly 60% of the way into the 8th week (I didn't feel like doing the continuous compounding calculation... sorry... but if they were continuously compounding, it'd happen slightly sooner). In a year, they'd have turned over about 36x their highest value observed. They'd get exempted 4x their highest value, which means they'd be taxed on 32x their highest value. If T were 1%, they'd be paying just under 1/3 of their whole portfolio in tax. They'd have to be superior to non HFT strategies by quite a bit if even a relatively benign HFT strategy is looking at turning over that much of their gain (and they'd have to be pretty confident they could do that doubling).
Of course, this will never pass, because it's much more fun to blame wall street. But it's food for thought.