pull down to refresh
182 sats \ 5 replies \ @Undisciplined 3h \ on: Research in Public #05: Does btc price affect SN behavior? Experiment failure! econ
Well explained and I'm glad stackers get to witness a snag in the process.
Some random thoughts:
- Interactive fixed effects: If you assume BTC price affects users and/or territories differently, then you can interact the BTC price with the other fixed effects. That should keep the time FE from eating all the variation in BTC prices.
- The rewards system is relative to what other stackers are doing, so there are incentives for zapping and posting that are disconnected from purchasing power of the sats spent. It makes more sense to have stackers maximizing their net rewards than only being price sensitive to the input costs. Do you see the rewards earned by stackers? I'm not sure what the right way to include that information would be, but there might be something there.
- Can you aggregate the value of the sats on a finer timescale? If you convert the sats to their dollar value at the time of posting, then the within week variation in bitcoin price would keep the time FE from absorbing the price information.
interactive fixed effects
Did you have any hypotheses about which users or territories might be more sensitive to price?
rewards system
I think I can back out some info about rewards. But more to the point, this reminds me that I need a model of zapping behavior. LIke, why do people even choose to zap?
finer timescale
One thought I had was not to bring the timescale lower (I think i would lose too much data), but to coarsen the fixed effects. Like, either use quarter fixed effects, or use a polynomial time trend, to allow weekly bitcoin price fluctuations to matter. This might be the most straightforward approach.
I'm also not a good time series econometrician (never took a class). So working with only time variation is not my forte.
reply
Did you have any hypotheses about which users or territories might be more sensitive to price?
I could come up with some ad hoc, but it shouldn't really matter. As a general hypothesis, I'd expect those who have been more active for a longer time to be less price sensitive and bitcoin related territories to be less price sensitive.
why do people even choose to zap?
That's partially answered in this post, at least in a way that no economist would doubt. From what I could tell, for most stackers, zapping more has earned them a net positive return. That's why I was on such a campaign to get stackers to zap more.
lose too much data
Why would you lose data converting to real-time dollar values? The fixed effects would still be weekly and granular btc price data must be available.
The difference should just be that you're averaging the dollar value of posting fees paid per territory, rather than multiplying the average bitcoin price for the week by the territory fee.
As I'm typing, I realize it wouldn't be easy to back out the direct impact of bitcoin price. You could still compare the sensitivity of posting behavior to bitcoin vs fiat denominated costs.
I'm not a time-series expert either and it was probably the part of econometrics I struggled with the most.
reply
Why would you lose data converting to real-time dollar values? The fixed effects would still be weekly and granular btc price data must be available.
You're right. I was thinking about it wrongly for a second there. I would be really surprised if daily fluctuations mattered even after controlling for week fixed effects. I should probably try it, but if that were to happen I'd probably think it was bot related.
That's partially answered in this post
This is why it's so tricky. If zapping actually leads to positive ROI denominated in sats, then higher bitcoin price should incentivize more zapping, not less. I do think there's some variation to be exploited regarding the return function between reward amount and zap amount.
Another tricky thing about this is that the reward system has changed multiple times throughout SN's life, which means I'd have to try and back out those periods.
reply
There was one month when the rewards were entirely based on zapping and another when they were entirely based on posting/commenting.
Maybe looking at those two months separately could tell us something about responsiveness to rewards for zapping.
reply
Good idea. I need to dig through k00b's announcements on reward changes and construct a history of regime switches.
reply