Well, the Mega Millions draw is getting close, the cash value is 94.2 million but the odds are 1:175 million. If you plug in the numbers, you get f=-492 but the contribution from the jackpot prize is only f=-1.9 hence the subsidiary prizes are way too small given the relative odds. I suspect that you would need a huge jackpot, much larger than would be necessary for a positive ROI before the Kelly's Criterion will tell you to buy instead of hedge because you would still have to offset the effect of all those smaller prizes which occur more often but I think I may not be attributing the costs of those subsidiaries correctly since those chances basically come free with the ticket if you've included that cost in evaluating for the jackpot (will have to work on that). Maybe that's why Princeton-Newton is a Hedge fund, there's probably more hedging opportunities then buying opportunities. Shannon also had a 50/50 portfolio rebalancing strategy that harvests volatility but there' the issue of the transaction costs but I'm sure the stockbroker would love it if you tried the method.
With horse racing, you're betting on inefficiencies in the bookmaker setting the odds that you're betting on hence you're assuming that your information about the odds are better than the bookie's. I wouldn't want to use Kelly's on the racetrack without identifying and tracking side information sources and analyzing them historically. That's unlikely with the racetrack as the sources are bound to be intermittent and you won't be able to access the same source for each race. However with the stock market, you can identify consistent sources such as rating companies, federal interest rate, unemployment rate, consumer index rate, US treasury bond rate, P/E ratio, sale volume, first derivative, second derivative and third derivative values etc. Indeed, most people use 0.5 as the probability for stock investment rather than track it as a function of side information, this is one of the reasons why people believe it's best for short term volatility but there is a logarithmic function relating current capital to previous capital and probability due to individual surprisal bits. I think it would be possible to come up with a reasonable approximation of the probability function but I don't know how to address the payout for stocks yet other than to set a increment value that I want to consider as sufficient for a transaction and to base the probability function on such movements with multiple surprisals to account for multiples of that price movement.
To bet a fraction of the Kelly bet is to allow for the possibility of unknown side information and is prudent with all Kelly bets. The thing is that with lottery, the probabilities are known exactly and the payouts are known with the exception of pari-mutuals, the problem with lotteries is that there are usually no inefficiencies in the prize structure to take advantage of.