Research Projects

tags: olympus

All the research projects will have the following form:

1. Experimental Design: A design of the experiments to be conducted
2. Parameter Sweeps: The variables which will be swept over to see sensitivities for system success
3. KPIs: Numeric scores for how well a given experiment performed in each parameter sweep, i.e. the ending price or what return an exploiter would get (a negative score would be better in this case)
4. Success Metrics: Boolean measures of success taken from the KPIs. For example, a success might be that an exploiter gets a negative return from an exploit.
5. Analysis of drivers of success: An analysis of what parameters drive success of the system through machine learning techniques, output would be information on how to drive system success against different scenarios or attack vectors

Soros Style Manipulation

This research will hinge on the idea of one large playing seeing an opportunity to try and push the price of OHM down quickly and possibly activate the mechanism for a re-adjustment and then profit off of it.

Experiment Design

• This will not be a simulation based experiment, but rather a deterministic experiment
• Given a past series of prices for OHM and a large whale with some principal amount P there will be a market manipulation attempt
• The steps of the experiment will be as follows:

1. A whale sees an opportunity to influence the RBS in the short term and makes a large sell, some percent S of P.
2. Time passes and the RBS mechanism is possibly triggered
3. The whale sells the rest (1-S) of P for possibly a profit
4. The incremental return from this manipulation versus selling all of P in step 2 is measured to see if the exploit returns a positive amount.

Parameter Sweep

The following will be the different parameter sweep variables to test:

1. The different past trading patterns
2. The amount of principal, P that the whale has
3. The sell amount S
4. The market reaction to a lowered price (increases in demand)

Success Metrics/KPIs

KPIs:

1. Return to Exploiter: A KPI of how much return the exploiter gets
2. Reserves Used: The amount of reserves used by the RBS

Success Metrics:

1. Negative Return: A boolean value for whether the exploit yields negative return

Panic Selling

This scenario will show the effects of panic selling and how it might influence price but also be guarded against.

Experiment Design

• There will be larger sell pressure at first which begins the panic selling
• There will be conditional selling for the system which means that at each timestep, there is the possibility that for any reason some participants might sell if the price is below a threshold.
• More formally, there will be a distribution P of the price at which a sell might occur, and a distribution N for the size of the sell at that epoch if the price is below. So, if the current price < sample drawn from P, a sell amount drawn randomly from distribution N is put into the system.
• If the RBS is stabilizing enough, the panic selling should subside and the system will get back to equilibrium. If not, there could be a death spiral of selling.

Parameter Sweep

The following will be the different parameter sweep variables to test:

1. Initial panic selling magnitude
2. The price to sell distribution, P
3. The amount to sell distribution, N
4. Parameterization of the RBS

Success Metrics/KPIs

KPIs:

1. Ending Price
2. Time to return to equilibrium: The time it takes before the system stabilizes (if it does) and is trading in equilibrium
3. Number of Interventions: The number of times the RBS had to intervene to stabilize
4. Reserves Used: The amount of reserves used by the RBS

Success Metrics:

1. Returned to Equilibrium: Whether the system returned to the equilibrium state
2. Time to Equilibrium Minimal: A boolean of whether the system stabilized before some C number of timesteps
3. Minimal Interventions: A boolean of whether there was only 1 or some C number of interventions before the system reached equilibrium againstem

Bond Volatility Research

Experiment Design

The experiment is conducted to simulate under different market conditions, with different OHM bond policies, how will it affect the volatility of the system.
The experiment will simulate 1 year span, with all OHM bonds created and expired within this time span. We expect to see some OHM price volatility especially at the time when a bond is sold and when a bond expires.

Parameter Sweep

The following will be the different parameter sweep variables to test:

1. Market volatility (model needs to change to accomodate this)
2. Total amount of OHM bond's face value
3. Distribution of OHM bonds (different starting dates; different tenors)
   • Every month release equal amount of bonds (3M, 6M, 9M, 1Y 2Y)
   • Front loaded monthly 25% month 1, 25% month 2, equal the rest
4. Differences on weighting of tenors
5. Parameterization of the RBS (reinstate window length)

Success Metrics/KPIs

1. Volatility:
   a. magnitude of price shock at the bond selling and expiration dates -> take percentage price changes, find absolute maximum % change, possibly only look at negative ones
   b. Standard deviation of percentage returns
   c. [optional] time for the shock to recover (e.g. return to 10% of the initial price)
   d. [optional] reserves spent for the shock to recover (due to RBS)

Success Metrics TBD after we meet on this

Optional time permitting: Maybe more useful things to look but requires additional modeling: Contrasting OHM bond with staking? Adding mechanisms to provide stability?