We also need an honest way to account for "projected yield" but I don't know how to do that because there are too many variables:

BitUSD that comes in from yield -- fees, short interest, whatever -- should go into a buffer fund.

Every block, some tiny percentage of the buffer fund goes into the "accrued yield fund."  Thus, with no future network income, the buffer fund would decay exponentially with some predictable half-life we can set.  Let's set it to one month.

A distribution event from the accrued yield fund will occur when the accrued yield fund has accumulated to 0.01% of the eligible BitUSD.  Eligible BitUSD are those that haven't moved since the last distribution event.  Iterating through every immobile BitUSD balance is a technical issue with this approach, but we can avoid it with a clever implementation [1].

Transparently showing projected yield can be based on the buffer fund's income.  You can also show a single APY number for the income in a year based on the assumption that the buffer fund's income on future days will be equal to its average per-day income over the last 30 days.  Then also show the range of what will happen if the income goes up or down by one standard deviation.

You could even show a graph of what the APY over time will be if the current funding level continues, if the income goes up or down one standard deviation, if the income doubles, halves, or goes to zero.

[1] We don't want to iterate through every immobile BitUSD balance on every distribution event.  So instead we merely record the block number at which the distribution event occurred, and move the 0.01% to a third fund, the distribution fund.  Then whenever a BitUSD balance requests its yield, you check how many distribution events occurred since it last moved (minus one, since it wasn't eligible for the first distribution event).  Then pay from the distribution fund 0.01% times the amount of BitUSD times the number of distribution events.

Besides solving the technical problem, this implementation also provides honest accounting of untouchable BitUSD, which are currently very hard to figure out.  The distribution fund can be immediately claimed by longs at any time.  The accrued yield fund and buffer fund are untouchable right now -- but the longs are assured that BitUSD will move from them into the distribution fund, in a timely manner, according to a simple, predictable, transparent algorithm.

Do we really need more than 400% reserve? How conservative do we want to get here? Like drltc pointed out, there is a balance here: the more conservatively the DAC operates the less value it can capture for useful purposes such as higher yields on BitAssets, or even payments to delegates/workers to grow the DAC in other ways.

I think this argument conclusively shows that having shorts pay interest will be better for the BitShares ecosystem than having shorts develop excessive collateralization.  bytemaster's comments in this thread and elsewhere lead me to believe that he disagrees; he thinks collateral ratio is more important.  (At https://bitsharestalk.org/index.php?topic=9532.0 I develop at great length another objection to excessive collateralization.)

I was trying to argue that there should be one variable and that variable should be interest.  Given bytemaster's strong support for collateralization ratio as the one variable, I thought interest on shorts would be likelier to be implemented if I offered the compromise of including both.  You rightly point out that this compromise will add complexity and make it harder for market participants to formulate strategy.

However, I worry that your function f is going to be way too complicated for market participants to reason about. How exactly is someone considering shorting supposed to decide on their strategy? They now have three variables to tweak: price, collateral ratio, and interest rates (I am assuming the market decides interest rates rather than having it be universally imposed by the network at any given time).

Dealing with the three-dimensional nature of the problem makes it difficult.

We could get rid of the "price" dimension by having two different types of shorts:  A fixed-price short with 2x collateral and no interest, and a "wall short" or "feed short" that moves with the feed but has variable collateral and interest.  Fixed-price shorts would only execute when there are no wall shorts, and they wouldn't really be useful when there's a lot of people wanting to short.

I was thinking of showing the f() score of each offer.  With one pretty graph of the total BitUSD of orders that would be ahead of you as a function of the interest rate assuming you kept the current leverage ratio, and another pretty graph of the total BitUSD of orders ahead of you as a function of the leverage ratio assuming you kept the current interest rate.

Then I realized that f() of each order changes with the price level.  Whether there's some f() that will result in this effect helping or hurting all orders equally and preserving their relative positions in the sequence is an interesting and non-trivial theoretical problem.  If you don't have an f() with this property, then the two pretty graphs are useless and misleading.  I think most f() will not have the necessary property unless you design it in.

Also, keeping some of the BitAssets from the yield fund in a reserve fund in case of a black swan event can be another mechanism, beyond collateral ratios, for the DAC to further reduce black swan risk (at the expense of lower yield rates) without needing to add complication to the shorts.

This is an idea I support, but I think bytemaster's plan for now is, if it becomes a problem, make a hardfork that can spend the "untouchable BitUSD" in the yield fund.  I agree that it should be considered low-priority given the market behavior and the other major changes they're working on, discussing, or have already rolled out.

I'll say the interest will be in "coins" in this post; I'll reveal whether interest should be BitUSD- or BTSX-denominated in the next post.

The simplest thing to do with the interest is to burn it as fees (if BTSX-denominated) or pay it as yield (if BitUSD-denominated).

We can do something more sophisticated, however:  If we want a large amount of capital tied up for a long time, we can use the interest to incentivize long-term investors to remove their capital from circulation.  We can let a long-term investor like Lenore make an order such as "I'll pay 1000 coins today, in order to receive 1030 coins in a year."  Ludwig might make a competing order that says, "I'll pay 1000 coins today, in order to receive 1025 coins in six months."

The network will convert these competing orders to a common scale -- APY.  If the network's only received, say, 25 coins in interest payments, then it will partially fill Lenore's order (which has a lower APY than Ludwig's order) and leave Ludwig's order unfilled.  To ease network load, maybe filling of partial orders should only happen once an hour or so.

This soaks up Lenore and Ludwig's spare capital without potentially setting up BitShares for its very own version of the late 2000's (decade) financial crisis.  And definitively addresses the issue of how we can still tie up large amounts of capital, even if we allow 2x positions to offer interest to out-compete higher-leverage positions (which is an issue I'm sure Agent86 would raise, based on his quote at the top of this thread).

Should the interest charged to shorts be BitUSD-denominated or BTSX-denominated?  The final question will be answered in the next post...

From the story in the previous post, the next natural step is to have a "BTSX money market" to match potential speculators like Alice, with potential creditors like Carrie.

The problem with this situation should be obvious to anyone who was paying attention during the US financial crisis at the end of the last decade.  Alice and other shorters' debts will be very highly rated, so a lot of people like Carrie will put a lot of money into them with an expectation that the principal will be preserved.

If BTSX takes a nose dive and covering Alice's short would dip into Carrie's $8000 principal, then Carrie and all the other creditors will start putting a ton of political pressure on the central bank to print BitUSD or BTSX to make them whole.  Regardless of the central bank's response or non-response, the crisis of confidence will jeopardize the entire system.

You can't stop Alice and Carrie from finding each other and making this business arrangement -- if the blockchain doesn't get built-in support for it, they'll just find a way to do it off-chain.

You can't stop Carrie from mentally turning "this deal is approved by all the credit rating agencies and has a very low probability of default" into "this deal has a zero probability of default."  Humans are notoriously poor when it comes to reasoning about very rare events.

The best way to prevent a financial crisis would be to have Alice require no capital from Carrie at all.  Allow positions like Alice's that want safe, but lower than top-of-book, leverage ratios to offer to pay interest directly to the network.

Should this interest be denominated in BTSX or BitUSD?  Who should get it?  How much interest should be charged?  What about the opening quote in the previous post?  Won't Carrie still want to do something with her capital?  Read more in the next post...

I feel there will be plenty of demand for BitUSD on a more collateralized network earning over 2% than a lower collateralized network.  I think long-term viability of BTSX depends upon how secure BitUSD holders are that they will not be left unbacked or illiquid.

The probability of black swan event decreases as a function of the collateral ratio.  I agree that long investors prefer a higher collateral ratio to a lower one, because it reduces the probability of a black swan event.  But long investors also like to receive interest!

If the current collateral level gives an annual black swan risk of 0.5% or once every 200 years, your line of thinking would imply long investors would rather reduce that risk to 0.001% through increased collateralization -- even if the alternative was cash interest from shorts.

This assertion is simply not credible.

You can always re short if you want so I don't understand what is the benefit for holding a short position for 1 year (even for the old shorts)...

If the price rose, then re-shorting the same number of BitUSD would require more collateral.

E.g. if price is at 30 BTSX / BitUSD today and you short 100 BitUSD.  Then BitUSD rises to 40 tomorrow, then drops to 20 in 6 months.  With one-year expiry, you have 3000 BTSX tied up for 6 months.  With one-month expiry, you have 3000 BTSX tied up for one month, then you have to re-short at 40 which ties up 4000 BTSX for the remaining 5 months (unless you re-short less than 100 BitUSD).

Yep.. we just discussed such a system... it would make it more expensive to be a market maker and I feel there will be plenty of demand for BitUSD on a more collateralized network earning over 2% than a lower collateralized network.  I think long-term viability of BTSX depends upon how secure BitUSD holders are that they will not be left unbacked or illiquid.

A market maker with e.g. 10x collateral would be able to offer a low or zero interest rate and still beat lower-collateral offers paying much higher interest rates.  Keep in mind also that the interest rate offers compete on is an annualized interest rate or APR, hence a market maker who expects to quickly cover, e.g. within one day (or less), will only pay one 1/365 (or less) of the APR.

I think that it is necessary to use interest rates to balance short-side supply of BitUSD with long-side demand for BitUSD.  A higher collateral ratio doesn't stimulate demand for BitUSD.

It is not sufficient for us to offer BitUSD to the masses.  We need to give them a compelling reason to buy it.  That might be a vibrant ecosystem of major merchants that accept BitUSD for goods and services -- but such an ecosystem does not currently exist.  The high yields that we would be able to offer to longs by charging shorts interest, would make BitUSD a much more attractive product than fee-based yield alone.

This is the exact implementation I had considered *if* we wanted shorts to pay interest to longs rather than posting additional collateral.

This would create a bidding system / auction system for setting the interest rate...

Why not do both?

Let every short offer specify both the collateral ratio, and interest rate.  You can increase your offer's sort priority by increasing either the collateral ratio or the interest rate.

In other words, just specify some hard-coded function f(collateral_ratio, interest_rate) with df/dx and df/dy both positive, and sort shorts according to their f() value.

Basically f() represents to what extent the network is willing to reward higher collateralization with an interest rate discount.