The crucial information is up and down ticks of the ES-mini futures at the Chicago Mercantile Exchange. These lead the S&P500 (SPY) index funding in NYC (by 4 milliseconds), and the ES-mini ticks up and down about once per second. This few bits of information carries huge value since much of the world's equity market responds to it. See https://arxiv.org/abs/1302.5966

Great link, thank you. Abstract in case it piques people's interest:

"High frequency trading has led to widespread efforts to reduce information propagation delays between physically distant exchanges. Using relativistically correct millisecond-resolution tick data, we document a 3-millisecond decrease in one-way communication time between the Chicago and New York areas that has occurred from April 27th, 2010 to August 17th, 2012. We attribute the first segment of this decline to the introduction of a latency-optimized fiber optic connection in late 2010. A second phase of latency decrease can be attributed to line-of-sight microwave networks, operating primarily in the 6-11 GHz region of the spectrum, licensed during 2011 and 2012. Using publicly available information, we estimate these networks’ latencies and bandwidths. We estimate the total infrastructure and 5-year operations costs associated with these latency improvements to exceed $500 million."

It's not amount of data but latency. A signal to buy is just one bit, but you need to get it before everyone else does.

Could an adversary send buy/sell fake signals and profit?

Given the low bandwidth I assume they can't use very sophisticated authentication mechanisms

It would be extremely trivial to transport a time-keyed, completely random one-time-pad out of band (e.g. over a commodity internet link/vpn) so long as there is sufficient difference in bandwidth, and then simply XOR the data with it.. good luck cracking that.

and that's just what a few seconds of thinking came up with.

How do you use a one-time pad on a single bit of data?

(Actually I see how that's possible, A XOR B, upticks are 1, etc.)

If the pad is coming via a side-channel then the speed increase is limited by that side-channel, you lose all your gains by waiting for the pad, then lose more by decrypting and verifying.

That aside my first thought was someone could jam the signal by blasting the receiver with noise; I wonder how they mitigate that.

A one-time pad is the right idea. I would send it on fiber at night before the markets open and then use bits from it as the market trades. A few MB per day would be plenty, given the dialup-like speeds of shortwave. That would go through a 10 gbps fiber in milliseconds.

The actual encryption of each bit just require an XOR on each end of the link. That's about half a nanosecond on a Xeon or one clock cycle on your FPGA.

A jammer wants to be close to the receiving antenna so he can use less power. Yet he wants to be far away so he's less likely to be caught. Willfully interfering with any any radio service is illegal in any regulatory domain AFAIK.

There is no need for a one-time pad. Any good stream cipher works just as well, same as for any other link. Message integrity is a fun question, since you don't want to spend the bits that you'd need for typical levels of security there; but that question is again independent of a one-time pad vs. normal crypto.

Jamming a licensed radio link vital to rich people who routinely deal with the government doesn't sound like a great idea to me.

Good point, a one-time pad is not needed.

Either way, the likelihood of loss between TX and RX means that you can't advance the cipher as bits arrive. I'm thinking you'd have to advance based on GPS-sync'd clocks. Given that each bit is at least 10's of microseconds, GPS sync is plenty good.