Interpretation of the liquidity tiers, a waterfall model, triggers, metrics and actions

masterOfDisaster · October 19, 2015, 2:21pm

Continuing the discussion from Tier 4 Fund Management Discussion:

The “Tier 4 Fund Management Discussion” would greatly benefit from an overview of the all liquidity tiers and an understanding how they interact with each other.
The “Finalized evolution of liquidity operations” defined the different tiers, but lacks the feedback loops between the different tiers.
The category “NuShares” might not be ideal, but I haven’t found a better one.

TL;DR

The current Nu liquidity operation still is rather based on gut feeling than on facts and thresholds. The sooner this gets changed, the earlier Nu can start to adjust the parameters based on experience.
Here’s a summary of the current liquidity model including an assessment/interpretation.
Funds get moved between tiers. Nu either does that or incentivizes it to happen.
Depending on the interface the friction is bigger or smaller, lead time is longer or shorter.
Thresholds that trigger action should be defined to make liquidity providing more reliable and to enable Nu with metrics.

A text wall will follow, but as there’s a lot of work to do in terms of developing fund handling on different tiers I went into greater detail.
The intention of this post is not to provide a complete solution; it’s rather trying to get the ball rolling.
Please join the discussion to define or refine the understanding of liquidity!

Summary and interpretation of the tiers

Tier 1
This liquidity is immediately available, being on the order book. Basically, the level of liquidity in a particular market should be matched but not exceeded by NuBot (annotation: and PyBot for ALP).

Interpretation:
Tier 1 is outside direct control of Nu. Nu controls tier 1 indirectly by providing incentives to put orders on sell and buy side.
That was once different when @KTm and @jmiller provided liquidity with funds that were owned by Nu, but has changed with the transition to the decentralized liquidity operation model.

Tier 2
This liquidity sits on exchanges but is not placed on order books and can be promoted to the order book (Tier 1) in a couple seconds. When Tier 1 liquidity is consumed and it can be verified that the corresponding trade in the real market has been successfully executed, NuBot promotes funds in Tier 2 to Tier 1 by placing orders.

Interpretation:
Tier 2 is outside direct control of Nu. Nu controls tier 2 and the promotion to tier 1 indirectly by providing incentives to put orders on sell and buy side. A shift of funds from tier 2 to tier 1 is incentivized with the compensation for tier 1 funds that is provided by Nu.
If Nu wants to have funds moved from tier 2 to tier 1 an increase of the compensation for tier 1 is the key.

Tier 3
This liquidity sits off exchange and is held by liquidity providers in wallets not exposed to counterparty risk or the risk of NuBot malfunction. It can be promoted to tier 2 when needed in minutes.

Interpretation:
Tier 3 is outside direct control of Nu. Nu controls tier 3 and the promotion to tier 2 indirectly by providing incentives to put orders on sell and buy side (which ultimately will move the funds from tier 3 to tier 1). A shift from tier 3 to tier 1 (via tier 2) is incentivized with the compensation for tier 1 provided by Nu. If compensation is provided for tier 2, Nu can incentivize moving funds from tier 3 to tier 2.

Remark:
Promoting tier 3 to tier 2 might take much longer than a few minutes. Depending on how long the confirmations take and how many confirmations are required by an exchange, it might take significantly longer than minutes.

Tier 4
This liquidity can be provided by custodians not dedicated to liquidity operations. A present example are the proceeds of NuShare sales. They are intended for operational and development expenses, but can be used to support the critical function of liquidity provision as needed. When these funds are used for liquidity, they are exchanged from one type of asset to another, but are still available for their original purpose, such as development. These funds can be promoted to tier 3 in hours and the cost is exchange rate risk.

Interpretation:
Tier 4 is in full control of Nu. Funds in tier 4 belong to Nu, are owned by Nu. Tier 4 is meant to be used for “operational and development expenses, but can be used to support the critical function of liquidity provision as needed”.

It might be helpful to distinguish between “colors” of tier 4 funds. In the end it might make things more complicated for little gain. Accounting wise it would be helpful to have separate instances of tier 4. In the end all is money that is owned by Nu and that gets managed for a purpose (or for purposes).
Development and operational expenses sometimes might be required to be paid in BTC. In fact a lot of development was paid with NBT and Nu doesn’t know how much of the NBT ended in the buy walls (which effectively made Nu pay with BTC).
Anyway, in case the funds on tier 4 buy side were depleted, BTC need to be obtained by selling NSR - no matter for what reason.
In the end it doesn’t matter whether tier 4 needs to be refilled because of development or because of keeping the peg.
Filling tier 4 is done by NSR holders vote. It needs to happen from a higher Tier.

Tier 5
This liquidity presents itself in a decentralized manner from the manipulation of interest rates (park rates). This liquidity is generally buy side, although technically it can be sell side in the case of lowering interest rates. This liquidity is available in days and the cost takes the form of interest paid.

Interpretation:
Tier 5 is in full control of Nu. Parking rates are adjusted by NSR holder vote.
Tier 5 is not a Tier like all others as it creates an inseparable, but time shifted effect on the buy and the sell side. Raised parking rates lower the sell side (increasing the ratio between buy and sell), because more NBT get parked. Later, when the NBT unpark, more NBT than were parked (and removed from the sell side) enter the market and might enter the sell side. The degree depends on the situation at the point of time the NBT unpark.

Tier 6
This liquidity takes the form of custodial grants for sell side and currency burning for buy side (presuming a currency burning motion is passed and implemented as appears likely). It takes a week or more to bring to market but has zero maintenance costs.

Interpretation:
Tier 6 creates NSR to have them sold for BTC to support the buy side or creates NBT to support the sell side. Creation of one or the other is necessary if the buffering capabilities of the lower tiers run dry. Tier 6 should create NBT or NSR and transfer them to tier 4 to (re)fill tier 4 funds or utilize tier 4 to bring them to market.
If NBT or NSR buybacks (to burn the retrieved asset) is intended, seeded auctions can be used to bypass tier 4.

Nu, the LPs and the tiers

The summary makes quite clear, that the lower tiers 1 to 3 are outside of direct control by Nu.
The funds in these tiers don’t belong to Nu (under normal circumstances). A special case is in the times NSR,BTC or NBT enter the market from a higher tier to get exchanged.
The tiers 4 to 6 are under direct control by Nu.
Tier 5 is a special case, because the NBT that get parked don’t belong to Nu. They are destroyed and created by protocol during parking and unparking. One could argue that the parked NBT create a temporal liability for Nu, Nu borrows NBT from the market to adjust the sell side, paying a compensation for that.

Waterfall effect

Adjusting volume on tier 1 by providing incentives to do so, siphons funds from higher layers. Increasing compensation for tier 1 will move money from tier 2 to tier 1. If the incentive is big enough and tier 2 is reduced, tier 3 funds will be moved from tier 3 to tier 2.
The thresholds for those movements are set by liquidity providers and can’t be adjusted by Nu.
All Nu can do to interfere with the waterfall effect on the levels 3 to 1 is providing an incentive to put money on tier 1 (and maybe tier 2).

Beginning on tier 4 it’s different. Nu does have the funds under direct control and depending on the market and liquidity situation funds from tier 4 can be injected into the market. The injection can happen on different tiers. The bigger the urgency the lower the tier which funds from tier 4 will interact with.
Slowest action would be to interact with tier 3; tier 2 would be faster; tier 1 has an immediate effect.

To be able to support both the buy and the sell side, tier 4 funds need to hold BTC (maybe a part in NSR to reduce the dependency from BTC) and NBT.
Whether the fund management of the different side is in different hands is something that needs to be defined as well as the amount of funds that shall be in tier 4 (in relation to the sum of funds in tier 1 to 3?) and the thresholds (and the velocity of the change?) that trigger action.

Metrics, triggers, actions

Tier 4, (5) or 6 involvement is required if tier 1, 2, 3 reach certain absolute values or ratios. As long as there are hundreds of thousands of NBT in the wild, absolute values are less important than ratios of buy and sell side.
If funds are running low in tier 1-3, they need to be refilled from tier 4-6.

Tier 4
The buy and sell side might have different thresholds to trigger action. Initially the same thresholds should be fine. That can be adjusted in following iterations.

Example:
tier 4 funds get activated if one side (sum of tier 1 to tier 3) has double the volume than the other. The balancing shall bring the sides in equilibrium.
As long as no side is at least double the other, nothing happens.
Please note that this is only an example; other examples will follow.

Remark:
The velocity of ratio change of buy and sell side could be included in the triggered action, but would make things more complicated. I’d rather consider that a next iteration step.

Tier 5
I have no clue how to intelligently include Tier 5.
As adjusting parking rates have a delayed effect on the liquidity, because it takes time until they are adjusted, I consider a continuous and early adjustment of parking interest more effective than adjusting the parking rates in case of emergency.

Example:
if the (average sum of tier 1 to tier 3 in the last time frame x) sell side liquidity is bigger than buy side, NSR holders are expected to raise parking interest above 0%.
And this is where I start getting clueless. How high to raise the parking rates depending on which circumstances?

Remark:
tier 5 can be used to reduce the sell side, increasing the ratio between buy and sell side. It can’t be used to increase buy side, obviously.

Tier 6
Assuming tier 6 is being used to fill/increase funds, it needs to be activated based on tier 4 status.
For the start it might be ok to have the same rules for both sides. As NBT on tier 4 sell side pose neither volatility nor default risk for Nu, they can be treated completely differently (q.v. FSRT).
Especially for the buy side it might be helpful to keep a ratio between funds on tier 4 and NBT in the wild, e.g. approximately 12% of NBT value need to be held in tier 4 buy side.
A corridor for the rate saves from frequent action.

Example:
the tier 4 buy side shall be kept on 12% with a corridor of +/- 3 percent points.
Once tier 4 buy side is below 9% NBT value in the wild (or a percentage of the sum of tier 1 to 3?), tier 4 buy side gets filled with funds from tier 6.
It’s strictly recommended to let a custodian grant pass for creating NSR, which will be exchanged for BTC to fill tier 4 buy side.
Once tier 4 buy side is above 15% of NBT value in the wild (or a percentage of the sum of tier 1 to 3?), tier 4 buy side moves funds to tier 6. BTC are handed over to tier 6 fund managers, who will exchange them to NSR and burn them; a shortcut is to have tier 4 managers obtain the NSR and burm them.

Remark:
using absolute values of NBT in the wild rather than the value of tier 1 to 3 makes it harder to game the system and should be preferred.

Conclusion

I know that this is a lot of stuff. Honestly I couldn’t have made a less elaborated version to make this vision clear.
I’m aware that there will be a lot of discussion, because interpretations will differ. I hope that in the end Nu will be closer to a “management framework” for dealing with funds.
The first draft won’t be perfect, but everything might be better than doing all based on gut feelings.

mhps · October 19, 2015, 3:19pm

nice summary. It’s still unclear how Tier 4 is injected to Tier 1-3, tiers that consist of a great number of LPs. That is the crux interface!

From reading your summay it seems that Tier 4 and Tier 5 can be the same thing only to react to Global Liquidity Imbalance (GLI; defined as global sell side liquidity minus the buy side; edit: then the difference is divided by the buy side ) changes at different frequency – Tier 4 reacts to fast spikes and dips of GLI, while Tier5 react to slow fluctuations of GLI. For a mathmatical example Tier 5 continuously reacts to the 7-day moving average of GLI, denoted as MA(GLI,7), Tier4 continuously reacts to GLI - MA(GLI,7).

masterOfDisaster · October 19, 2015, 3:40pm

Thank you. Like I said, I didn’t intend to provide a complete solution. I rather wanted to get the ball rolling
It would very well be possible to let tier 4 fund managers conduct the seeded auctions once they are available.
With thresholds like “one side is double the other” different actions could be triggered;
it would be clear which side to seed depending on a condition of the liquidity situation.
Or what about this: in “seeded auctions” seed one side if the other (NSR would take the place of BTC) side is above 1.5 times the other, but less than 2 times.
Above a ratio of 2 times placing orders on exchanges (tier 1 injection) could be the triggered action for tier 4 managers.
…not saying that 1.5 or 2 would be the ideal ratios…

This is still a brainstorming. It needs to be shaped later

It might be easier to think about interfaces if we have an understanding of the kind of action that shall be triggered.

zoro · December 3, 2015, 6:20pm

At last, i finished reading
I will give you my point of view as an LP.

I consider T1 and T2 as a combined tier. Meaning that funds (of both tiers) reside in the same exchange
and are used automatically by the bot in a “constant reward” pool. In a “fixed cost” pool, T2 doesn’t exist,
unless the new nubot has functionality to play with T1 and T2 independently.
Anyway, for an LP operating only in one pool, T2 cannot be used to balance T1 of other pools, thus we cannot say
that T2 can be used to balance T1 in all cases (that is why i consider them combined).

About T3, i believe T3 has 2 roles.(2 sublayers)
T3.1 is the “personal” tier of an LP used as he/she likes to send funds to T1-T2 of any exchange he/she feel like.
NU cannot now about T3.1 unless the new nubot has functionality to play with, and it can actually report back T3.1 balances.
T3.2 is the sublayer closer to T4 that NU has full knowledge of it. It is the tier that interacts with T4 and T1+T2 to make
fund transfers faster with 24/7 availability.
It could be managed by an elected trusted T3 custodian with single signature addresses
or a small group of FLOT members with 2 out of 3 multisignature addresses, or both, why not?

Thus my vision of tier flow (from T1 to T4) is like below:

T1+T2 — T3.2 —T4 —…
… \ … /
…T3.1

That’s it, as an LP i don’t go beyond T4, sorry

masterOfDisaster · December 4, 2015, 10:00am

This depends on how you define or interpret T2.
T2 funds might be interpreted as fund being on the order book, but outside the spread for which compensation is paid.

Liquidity providers who do that forfeit compensation of those funds, but get them traded for a nice uplift if they can sell them in times in which the T1 funds (the ones inside the defined max. spread)are temporarily depleted.

NuBot supports a parametric order book.
The funds on T1 can be limited, the rest of the funds is put in orders at more expensive exchange rates.
This is the supported and intended operation model of NuBot, because it decreases the risk for LPs to fall victim to hedging with a lot of money, leading to decreased costs and cheaper liquidity operations.

If I understand the operation of NuBot correctly, funds on the closest spread (T1) don’t get replaced if consumed by matched orders, until the price moves and the oders get cancelled and placed anew.

If the BTC price swings in any direction, the amount of funds that can be traded in a short period of time for a close spread is limited. Hedgers might choose to pay a premium for trading beyond T1 or need to wait until the price moves to find T1 funds again.

I hear what you say. I fully agree to the function of T3.2. But I’d explain it in a slightly different way.
T3 sell side (T3 buy side has millions of BTC ready ) custodians hold NBT that are their property.
They verify that they hold the NBT and pledge to use them for T1 (and T2) operations.
They get compensated for having NBT “locked” (pledged to liquidity operations).
This compensation is lower than T1-2 compensation, because no exchange default risk applies, but needs to be big enough to make people hold the NBT.

I’ve thought some time about the accounting of the compensation.
I still believe that one of the easiest ways to do that is to make this via custodial grants by people who run NuBots.
The T1-2 funds get reported by NuBot.
T3 funds can be read on the block chain.
The total of T1-3 is more or less constant and only prone to BTC volatility when being traded hence and forth.

The incentive for the custodians to move funds from T3 to T1 is the increased compensation for a sell wall that has run low. This of course works even better with the fixed compensation scheme, because it amplifies the compensation effect of providing liquidity in low walls. With fixed rate it only provides an increased incentive, if the side drops below the compensation threshold.

FLOT is only doing business with these custodians, but not with random people.
That limits the involvement of the FLOT in liquidity operations and ties T4 to T3.
Plus it can allow (by NSR holder vote, depending on the terms of the custodial grants) to have the FLOT send NBT before they received BTC from the custodian. This speeds up the refilling of T3 in case the funds run really low there.
It might not play a big role, but can make the difference between the FLOT doing business with T3 and the FLOT beind required to use NBT entry gateways to directly inject NBT in T1 - which requires to have such a gateway at the exchange with the low sell side!

T3 custodians should trade with each other to balance sides on different exchanges as long as the overall T1(-3) sides are balanced.
The easiest way to do that is by transferring funds to another exchange and trade there.
If on one exchange there is demand for NBT, while another has sufficient NBT or NBT in excess, send BTC there, trade them for NBT and send the NBT back to where they are low.

T4/FLOT would only be involved if one side is overall low or in emergencies.

Nagalim · December 4, 2015, 2:05pm

What’s the incentive for a custodian to report their T3 funds? What guarantee do we have that those funds will actually end up being used in a peg break scenario? If those funds have no way of being accessed directly, but only at the custodian’s whim, how can we even call them T3? I don’t think there is an ALP solution for T3, though NuLagoon is showing the MLP solution. I propose a different kind of T3 custodian, one that uses Nu’s funds so they have every incentive to report and follow through in peg breaks, over in the T3 custodian thread.

masterOfDisaster · December 4, 2015, 2:23pm

To receive a compensation.

No guarantee, but the incentive for the T3 custodians to put funds in the shrinking walls to earn more reward (in percent) for their funds (only true for fixed compensation).
This makes a peg break scenario less likely. They have an incentive to fill T1 early.

The direct control over funds in the decentralized liquidity operations model ends at T4. T1-3 can only be steered by providing incentives.

That’s a viable solution, but leads away from the (completely) decentralized liquidity operations model.
At another place I’ve already written that some small steps away from the decentralized liquidity operations model might be good:

Nagalim · December 4, 2015, 2:30pm

If you’re compensating custodians for reporting T3 then you’re running some kind of ALP-type operation. So what prevents Satoshi from submitting his kajillion btc as buy side T3, with absolutely no intention of ever bringing them to market. Are we going to pay him for holding his own btc in his own wallet and having nothing to do with Nu?

masterOfDisaster · December 4, 2015, 2:52pm

Without any compensation, you won’t find people doing it.

The terms of the grant limit the amount of BTC or NBT.
T3 sell side custodians would be limited anyway.

Nu pays elected T3 custodians based on terms. People who get elected are aware of the T1 compensation. They will happily move funds from T3 to T1 if the compensation is sufficient

Nagalim · December 4, 2015, 3:29pm

A contract that includes a profit mechanism that is dependent on reporting also works (like what I’m proposing, in which T3 custodians are paid based on how many funds they sell or buy). If we pay based on the risk, holding funds in a wallet off exchange has 0 risk, so we should be awarding nothing. If we pay based on the work done, reporting can be made automatic and simple, so a basic operator fee would be all that would be required. But then, what motivates anyone to lose their anonymity by actually putting any funds in the reported wallets? Clearly, the best thing I could do would be to report a ton of wallets that all have like $1 of funds in them and collect a ton of operator fees. So the shareholders will be responsible for holding the custodians accountable for not having enough money in their wallets. I just think that’s going to be exhausting for shareholders.

Maybe. Exchange risk is pretty big. Ignoring opportunity cost, I’d happily hold a large amount of NBT in a wallet somewhere if Nu is paying me LP rates, especially if I never have to risk bringing it to exchange. That’s like 100x better than parking because I can use the money whenever I want.

mhps · December 4, 2015, 4:01pm

how about a T3 custodian must put the fund in a 2x2 multisig address which he/she and another party have to both sign and both pledge that will only transfer fund to a T1/2 address. T3 Liquidity is reported from the multisig address. The T3 custodian can’t spend the T3 fund on anything else. To make it pointless not to ever use T3 fund, compensation to T3 is only paid when it is transfered to a T1/2 address. The amount is proportional to how long the fund is in the T3 address and the transfered amount.

Nagalim · December 4, 2015, 5:23pm

This is great, each T3 custodian gets a watchdog. Let me think about it some more, but this seems like a solution.

Nagalim · December 5, 2015, 3:18pm

@mhps It seems to me we don’t even need the watchdog. All we need is the T3 to blackball themselves as a recipient of the funds. As anyone can fake an account, that means the T3 custodian is given a list of addresses corresponding to other trusted community members such that the custodian and that community member are unlikely to collude. Then, all Nu has to do is look through the custodian’s Txn history (written on the blockchain) and make sure all receiving addresses are valid T1 custodians. Then they pay them based on the time the liquidity was provided for after the fact.

Others, please try to attack this. Pretend you are the T3 custodian with malicious intent, how would you fraud Nu?

zoro · December 5, 2015, 3:25pm

How can you ensure validity of T1 ALP custodians?

Nagalim · December 5, 2015, 3:29pm

Well, any MLPs are clearly good (MoD or henry). Then the shareholders can elect some others like you. The main idea is to just make sure the T3 custodian and the T1 custodian aren’t working together.

zoro · December 5, 2015, 3:33pm

So i cannot be T3 and T1 custodian at the same time?

Nagalim · December 5, 2015, 3:35pm

You can, but you cannot make Txns with yourself from T3 to T1. The reason for this is a little convoluted. Basically, the T3 custodian has incentive to favor themselves and blackball all others if we let them trade with themselves. If the T3 custodian is not allowed to transact with anyone they can collude with, they are forced to treat all valid T1 recipients as equals.

mhps · December 8, 2015, 2:28am

I don’t see why this is a problem. If a T3 custodian gives himself T3 funds to put on his T1 bot and do all the reporting, and nu knows the fund is used to support the peg, everything is fine to me.

Nagalim · December 8, 2015, 1:07pm

What if the T3 custodian deals with no one but themselves, they send the money to themselves and put it up on T1 just long enough to get Nu to pay for the weeks of T3 service, then they take the money off T1 and walk away, getting paid exorbitant rates for like a day of actual liquidity support.

We would need to contractualize the T1 service as well if we let them blackball everyone but themselves. The whole thing works way better and simpler if you force the T3 custodian to follow a white list that they are not on. It keeps them from colluding.

mhps · December 8, 2015, 1:32pm

Then the question is how to fairly calculate T3’s reward period. What you said could happen to any T1 LPs. How/when do you think fund should flow back to T3 from T1-2? How about reward T3 for the pereiod when its fund is in T1-2? How about making T3 a loan from T3 owners to T1-3 LPs? Maybe they can settle on a price themselves?