Under the hood of crypto price aggregators

Cryptocurrency price aggregators are black boxes — no one really knows how they calculate average prices. However available info is enough to understand that they do it incorrectly. We uncover the problems and propose a professional pricing tool that is 100% transparent. It is frustrating that cryptocurrency movement relies on completely opaque pricing data. How is it possible to take any decision based on data from a black box? Over a year ago we started developing our own crypto price aggregator to fix this problem. We solved numerous methodological and technical issues during this year — now we want to share our findings: THE SILICOIN Aggregators pool trading pairs from all exchanges into one bucket to calculate average prices. This approach distorts prices. We should first find prices of coins on each exchange and then find averages across exchanges (see Section 2 of this article) There are many ways to find USD price of coins that do not have direct trading to US dollar. Most of them are inappropriate. We discuss available options and explain why step by step pricing is optimal. We discuss these findings below. They gave birth to the pricing methodology used by . Our methodology is described in maximum detail in . THE SILICOIN Section 4 This material is lengthy. However, we are sure that it is worth investing 15 minutes to understand once and forever how data that we look at every day is calculated. If you have questions about this article, you are welcome to ask them in the comments. Contents Grey areas and problems of CoinMarketCap’s methodology Why pooling all trading pairs in one bucket is wrong 3.1. 3.2. 3.3. 3.4. How CMC may be calculating reference prices Option 1 — step by step propagation Option 2 — iteratively recalculate prices for skipped pairs Option 3 — use previously calculated prices as reference Which option CMC is likely to use? 4.1. 4.2. 4.3. 4.4. 4.5. THE SILICOIN coin price calculation methodology Methodology summary Methodology results Detailed methodology Why use the most liquid trading pair to calculate coin price on exchanges How to observe all inputs used for calculations Conclusion 1. Grey areas and problems of CoinMarketCap’s methodology There are major unknowns and problems in CoinMarketCap’s methodology. We are talking about CMC not because its methodology is particularly bad, but because it is a market standard. All crypto price aggregators that we know use some variation of CoinMarketCap’s black box methodology. At first sight CMC’s methodology seems straightforward: Receive data for all trading pairs from exchanges, apply cleaning and verification algorithm Step 1. Pool all trading pairs and find USD price and volume for each pair Step 2. Calculate volume weighted average price of coins across all trading pairs Step 3. Methodology for Step 2 is as follows: “The price for each individual market pair is calculated by taking the unconverted price reported directly from the exchange and converting it to USD using CoinMarketCap’s existing reference prices ” ** **As we show in below, this approach yields ambiguous and abstract prices There are two major issues with CMC’s methodology:1. It is incorrect to aggregate trading pairs from all exchanges into one pool. Section 2 ** **We don’t even know what trading pairs are used to price Bitcoin.Calculation options are described in 2. How are reference prices used in Step 2 calculated? Section 3 On top of that we have no idea about what “data cleaning and verification algorithms” are used. It is ridiculous, but it is the least important problem. 2. Why pooling all trading pairs in one bucket is wrong As discussed above CoinMarketCap pools trading pairs from all exchanges and calculates averages across all of them regardless of what exchange a pair belongs to. From what we could find about methodologies of other crypto price data providers, they also pool trading pairs ( , , ). example 1 — Nomics example 2 — CoinPaprika example 3 — Bitwise Approach of pooling all trading pairs is flawed. It outputs incorrect prices. The alternative to pooling all trading pairs is to group them for each exchange. Every coin traded on an exchange should receive a unique price. : Bitcoin costs $4,000 on Binance. : Bitcoin costs $3,990 on BTC/USD and $4,010 on ETH/BTC trading pairs on Binance. Correct Incorrect A bit of theory: If there are no market frictions, do not exist. That means that assets have the same price across all markets where they are traded. arbitrage opportunities There are severe market frictions when moving funds from one exchange to another. That is why it makes sense to calculate average price of cryptocurrencies . across exchanges On the contrary, within one exchange frictions are minimal. Therefore, if we take trading pairs with some coin on a particular exchange — price of this coin in each pair should be the same. Consider an example: Assume there are 2 trading pairs on exchange “XYZ”: MKR/BTC=0.092 and MKR/ETH=3.60 First we calculate USD price of BTC and ETH using trading pairs from other exchanges where they are traded to fiat. Let BTC=$3,400 and ETH=$90 (these are infamous — to be discussed ) reference prices below Let’s calculate Maker’s price in both pairs. From MKR/BTC price is $3,400*0.092= , from MKR/ETH price is $90*3.60= $312.8 $324.0 We arrived at two different prices for the same crypto on almost frictionless market. In reality it is not possible! Traders immediately discover any deviations and eliminate them. Why that happened? We introduced 2 outside prices to a single exchange. Traders on this exchange value BTC relative to ETH differently than an average trader that we found by calculating BTC and ETH reference prices . It should be derived from the most liquid pair that has USD price already established To eliminate this problem we should find one best price for Maker on this exchange Assume trading volume of MKR/BTC pair is higher than of MKR/ETH pair. Than we should take BTC price of $3,400. MKR=$3,400*0.092= , ETH=$312.8/3.6= $312.8 $86.89 on exchange “XYZ” This problem can be easily illustrated. As per CoinMarketCap, Tether (USDT) price on BITTREX differs by over 3% between trading pairs, while transaction fee is only 0.25% and trading volume is substantial. That is clearly not possible — such deviation would be immediately exploited by arbitrageurs. Source: . Price difference may fluctuate CoinMarketCap 3. How CMC may be calculating reference prices When pooling all trading pairs are the major input — they have huge effect on prices of all other coins. reference prices Although we’ve already showed that pooling is wrong, let’s discuss how may be calculated. That helps in grasping the intricacies of pricing cryptos. reference prices 3.1. Option 1 — step by step propagation First calculate volume weighted average USD price of Bitcoin across BTC/fiat trading pairs (Bitcoin has the largest number of pairs to fiat currencies) Step 1. Calculate USD prices for trading pairs to BTC or fiat (coin/BTC or coin/fiat). BTC’s price is the . Then calculate volume weighted average prices for these coins. Not all coins have trading to BTC or fiat. Therefore, not all coins are priced so far Step 2. reference price Calculate USD prices for coins that have trading pairs to coins priced in Step 2. Coins that were priced in provide in Step 3. Step 2 reference prices Step 3 Continue such iterations until all cryptos receive USD price Step 4. Note that by far not all BTC trading pairs will be used to price Bitcoin under this option. Pair sample will be limited for all coins except those priced in the last step. Assume we pooled the following pairs from all exchanges: BTC/USD, ETH/BTC, ETH/USD, LTC/BTC, LTC/USD, LTC/ETH, MKR/ETH, MKR/LTC, ADA/MKR Consider an example for this procedure: Find BTC price in USD using pair BTC/USD Find ETH and LTC prices in USD using ETH/BTC, ETH/USD and LTC/BTC, LTC/USD Find MKR price using MKR/ETH, MKR/LTC Find ADA price using ADA/MKR Option 1 — step by step propagation Pair LTC/ETH is not used at all. Pairs ETH/BTC, LTC/BTC not used to price Bitcoin. Pairs LTC/ETH, MKR/LTC not used to price Litecoin, etc. 3.2. Option 2 — iteratively recalculate prices for skipped pairs Similar to Step 1 of option 1 Step 1. Similar to Step 2 of option 1 Step 2. Recalculate prices of coins from Step 2 using all available pairs (only BTC price is fixed), find Step 3. updated reference prices Calculate prices of remaining coins Step 4. Recalculate again Step 5. Continue such iterations until all cryptos receive USD price and all pairs are used Step 6. Under Option 1 many pairs are not accounted for. Option 2 tries to fix the problem. Let’s consider the example described above using Option 2 approach: Similar to Step 1 of option 1 Similar to Step 2 of option 1 Recalculate ETH and LTC price using LTC/ETH pair not used in Step 2. One can fit ETH dollar price found in Step 2 into the equation and get LTC dollar price and LTC/ETH dollar volume. Now we can recalculate LTC average dollar price using three pairs. Then one can fit LTC dollar price found in Step 2 (or even a recently recalculated LTC price) into LTC/ETH equation and get ETH dollar price and new trading volume from this pair Price MKR Recalculate LTC and ETH price again Price ADA Option 2 — iteratively recalculate prices for skipped pairs In Step 3 we generated a wonderful randomness by arriving at two different USD trading volumes for LTC/ETH trading pair which is colored red. 1st obtained by using ETH to find dollar values of LTC/ETH pair, 2nd by using LTC to find dollar values of this pair. We can make infinite number of iterations. After each iteration we will arrive at different prices. We can pray for convergence, but even if that happens resulting prices are…whatever but market average. We can also modify options 1 and 2 by first fixing USD prices for several top-tier coins (not only BTC) or try iterating calculations in any step for any number of times. However, we will always either omit some pairs or generate randomness by iterating. 3.3. Option 3 — use previously calculated prices as reference This is the worst option because prices become sticky and change slowly. Assume we have only two pairs: BTC/USD and ETH/BTC. Let previous average price of BTC=$4,000, previous average price of ETH=$100. Current BTC/USD quote is , volume is $9,000 and ETH/BTC quote is 0.027 BTC, volume is 1 BTC. $3,000 From ETH/BTC=0.027 → BTC=$100/0.027=3,703.7 Then new volume weighted average BTC price is (3,000*9,000+3,703.7*(1*3,703.7))/(9,000+1*3,703.7)= Average ETH price is 0.027*4,000=$108 $3,205.16 This algorithm outputs BTC price of while it should be . Outputs will eventually converge to actual prices, but that will take a lot of time. $3,205.16, $3,000 3.4. Which option CMC is likely to use We tried to reverse engineer CMC’s average BTC price from the data provided in the . We experimented with several approaches: markets section of CMC’s Bitcoin page Include all markets except those marked as excluded from average price Include only fiat markets Combinations of these None of the approaches gave us average BTC price that we saw on CoinMarketCap. We did the same with a smaller coin — Maker. In this case average price appeared to be just volume weighted average as promised. . Thus, CMC’s approach looks like some variation of Option 2 . Please try making your own calculations and let us know if any remarkable results. Google spreadsheet with our calculations can be found here 4. THE SILICOIN coin price calculation methodology (or how to calculate average coin prices properly) Below we lay out methodology for calculating average prices that cures the problems described in and above. Section 2 Section 3 This methodology is 100% transparent, robust and uses real transaction prices. Input data for calculations is available on — you can always verify our calculations. For example, see Exchanges section of for details of Ethereum average price calculation. THE SILICOIN Ethereum page 4.1. Methodology summary **For each exchange calculate price of each coin traded on this exchange**1.1. Price of a coin on particular exchange is established from the best trading pair, i.e. pair that has minimum number of steps to fiat and maximum transaction volume1.2. For exchanges that do not have fiat trading use one top-tier coin to establish a link to fiat Calculate total trading volume of a coin on particular exchange as sum of volumes across all trading pairs with this coin on this exchange Calculate volume weighted average coin prices across all exchanges 4.2. Methodology results Benefits of this methodology are clearly visible. Let’s compare Tether (USDT) prices on and on . CoinMarketCap THE SILICOIN There are only 2 exchanges that offer liquid USDT trading to USD — Kraken and Bittrex. USDT/USD price on these exchanges indicate true market Tether’s price. CMC arrives at inflated USDT prices on most trading pairs except USDT/USD markets that are priced directly from fiat. See screenshots below. As of 18 Dec 2018 Source: CoinMarketCap **With our methodology we arrive at stable prices across all exchanges.**Moreover, users can always verify that average price of a cryptocurrency is calculated by multiplying Price by Adjusted share (see below for details). full methodology As of 18 Dec 2018 Source: THE SILICOIN 4.3. Detailed methodology We encourage you to be patient and invest your time in reviewing detailed methodology. Step 1: Retrieve raw data from exchanges From each exchange receive last price and 24-hour trading volume for available trading pairs (i.e. BTC/USD, XRP/ETH, etc.). Step 2: Find USD price and trading volume of coins on each exchange Procedure of finding USD price of a coin on an exchange depends on whether this exchange offers trading to fiat or not. This part is important, please pay attention to it. We provide calculation example below for more clarity. 2.1. If liquid fiat trading is available on the exchange — price all coins directly Find USD prices of coins traded directly to fiat. Fiat currencies other than USD are converted to USD using data 2.1.1. Open Exchange Rates In case there are several trading pairs that can establish coin’s USD price in one step, pair with the highest trading volume is selected. See example below Find USD prices of coins traded to coins priced at a previous step but not traded to fiat 2.1.2. Such step by step iterations continue until all coins on the exchange receive USD price 2.1.3. base coin 2.2. If no liquid trading to fiat on the exchange is available use We use USD price of an applicable to find USD prices of other coins on this exchange 2.2.1. base coin are top-tier coins that are extensively traded to fiat on other exchanges. are used to find USD price on exchanges with no fiat trading — serve instead of fiat currency in this case. Base coins Base coins base coins is selected for each exchange separately — coin with the largest trading volume is selected. So far Bitcoin offers maximum trading volume on all exchanges with no fiat trading that are integrated by . Base coin THE SILICOIN For example, Bitcoin serves as on Binance. base coin Only one may be selected for an exchange to assure price homogeneity as explained in . base coin Section 2 of this article Same algorithm as described in 2.1. is applied to find most efficient path to for other coins 2.2.1. base coin 2.3. Calculate total trading volume for each coin on the selected exchange Calculation example for Step 2 of THE SILICOIN methodology Assume that exchange “XYZ” has the following trading pairs: Trading data retrieved from imaginary exchange “XYZ”. USD/EUR retrieved from Open Exchange Rates There are 3 pairs to fiat: BTC/USD, BTC/EUR, ETH/USD. Therefore, we can price BTC and ETH at this step 1. Price coins traded directly to fiat ( item 2.1.1. of detailed methodology ) For BTC we have 2 pairs to fiat. Dollar volume of BTC/USD=$15,000,000; of BTC/EUR=10,000,000/0.88=$11,363,363. We select the pair with higher volume → BTC/USD. Last price of BTC/USD= → that is BTC price on this exchange $3,200 For ETH we have only 1 pair to fiat. Therefore, ETH costs on this exchange $85 We explain why we use the most liquid pair approach instead of averaging in . Section 4.4 below 2. Price coins that are not traded to fiat but traded to coins priced previously ( item 2.1.2. ) For LTC we have 2 applicable pairs: LTC/BTC and LTC/ETH. Dollar volume of LTC/BTC=160*$3,200=$512,000, of LTC/ETH=3,500*$85=$297,500. We use pair with higher volume. Therefore, LTC costs 0.008*$3,200= $25.6 3. Calculate total USD trading volume for all coins ( item 2.3. ) For BTC we have 3 pairs with following volumes: BTC/USD=$15mn, BTC/EUR=$11.36mn, LTC/BTC=$0.512mn. Total BTC volume is $26,872mn For ETH we have 2 pairs with following volumes: ETH/USD=$6mn, LTC/ETH=$0.2975mn. Total ETH volume is $6.2975mn For LTC we have two pairs with following volumes: LTC/BTC=$0.512mn, LTC/ETH=$0.2975mn. Total LTC volume is $0.8095mn Illustration for calculation example: Illustration for calculation example Step 3: Calculate market average price of coins We combine data from all exchanges to find market average price of coins. Exchanges may be excluded from calculating average price for some coins under rules laid out below. 3.1. Market average price of a coin calculated as volume weighted average price (VWAP) across exchanges integrated by THE SILICOIN and where this coin is traded The VWAP formula: — price of a coin on as determined in Step 2, — total trading volume of a coin on as determined in Step 2, — weight of an after adjustments, — adjustments applied to raw weight of , — raw weight of an ,**Sum(V_i_)** — total trading volume of a coin across all integrated exchanges that are not excluded. In other words, it is sum of individual V_i_ — exchange integrated by THE SILICOIN and not excluded from calculation of totals for selected coin Where: P_i_ exchange i V_i_ exchange i WA_i_ exchange i adj_i_ exchange i W_i_ exchange i Exchange i 3.2. Exchange may be partially or fully excluded from calculation of VWAP and Sum(Vi) of a coin under following conditions if coin serves as a on this exchange (WA_i_ in the formula above is set to 0) 3.2.1. Automated full exclusion from VWAP base coin if 24-hour trading volume increased over 300% day-over-day and exceeds $100k (exchange is excluded from Sum(V_i_), therefore W_i_ and WA_i_ are set to 0) 3.2.2. Automated full exclusion from both VWAP and Sum(V_i_) if price of a coin on an exchange has large deviation from VWAP 3.2.3. Automated partial or full exclusion from VWAP _In this case we adjust weight of an exchange ( WA_i= WA_i= W_i in the formula above). Adjustment applies linearly to price deviations from 50% to 100%. For price deviation of 80% W_i*(1-(80%-50%)/(100%-50%))= WA_i=0, for price deviations below 50% W_i*0.6. For price deviations above 100% W_i Automated exclusion algorithms are set to cut off extreme deviations. Smaller deviations may be detected manually and exchanges may be excluded from VWAP or from both VWAP and Sum(V_i_) for some coins 3.2.4. Manual full exclusion. Note that WA_i_ are always rescaled to sum up to 100% **Note on 3.2.2.**This mechanism protects from exchanges sending us erroneous volume and distorting the price. $100k threshold is set in order not to cut off coins with low trading volume that tend to exhibit high volume volatility. **Note on 3.2.3.**We first calculate VWAP using W_i_. Then we compare prices on each exchange with VWAP and implement adjustments to W_i_ according to the formula. Next we recalculate VWAP using WA_i_. Exchanges that dominate by trading volume are unlikely to be penalized. This procedure occurs only once. 4.4. Why use the most liquid trading pair to calculate coin price on an exchange As described in we use trading pair with the highest volume to determine price of a coin on an exchange at each step. Step 2 of the methodology Alternatively, we can use average price across all pairs available in one step (both BTC/USD and BTC/EUR in the ). example above Below we explain why using the most liquid pair is better than averaging. Price of a coin within one exchange may vary across trading pairs for the following reasons: Bid-ask spread (Arbitrageurs may be unable to eliminate price deviations because of fees. Such deviations move stochastically and should have mathematical expectation of zero) Transaction fees No trading These inefficiencies obscure the true price, i.e. the price at which users can transact on the selected exchange. The most liquid pair provides more precise price than averaging: By accounting for lower volume pairs we will widen the spread 1. Top volume pair has the narrowest bid-ask spread We use the last price for calculations — it may be on both buy or sell side of the spread. Averaging may theoretically yield the mid price. However, during buy or sell markets, last price for each pair will be on the same side of the bid-ask interval most of the time. Therefore, averaging will widen the spread On a lower volume pairs, off-market price may hold for some time. Assume arbitrageurs see mispricing on low volume pair and try to eliminate it, but no one is buying/selling on the other side. The highest volume pair is less prone to such distortions 2. Lower volume pairs are more likely to exhibit inefficiencies 3. It is just more straightforward and people are more likely to be able to transact at prices of the most liquid pairs 4.5. How to observe all inputs used for calculation of average prices Contribution of each exchange to average price and total trading volume of a coin: You would find exchange weights for each coin in “Exchanges” section of the dedicated page of selected coin. For example, or Bitcoin page Ethereum page We provide W_i_ from the in “Market share” column of the table and WA_i_ in “Adjusted share” column VWAP formula If Market share = 0, then exchange is excluded from both VWAP and Sum(V_i_) calculation for selected coin (footnotes 3 or 5, details below) If Adjusted share = 0, then exchange is included in Sum(V_i_) but excluded from VWAP for selected coin (footnotes 1, 2 or 4) 2. Adjustments and outlier cleaning mechanisms applied to exchanges: If any conditions described in apply to an exchange, a footnote will appear near the name of this exchange in the “Exchanges” section of the coin page. Below is the legend: Section 3.2. of the methodology — fully included in calculations of aggregates — Base coin ( ) — Price outlier ( ) — Volume outlier ( ) — Manually excluded from average price ( ) — Manually excluded from average price and total volume ( ) No footnote 1 see item 3.2.1. of the methodology 2 see 3.2.3. 3 see 3.2.2. 4 see 3.2.4. 5 see 3.2.4. 3. What trading pair was used to establish USD price of a coin on each exchange In “Exchanges” section of any you can find “Pricing pair” column that contains trading pairs used to establish USD price of the selected coin on each exchange and a link to historical trading data of this pair coin page in Pricing pair column means that selected coin was used as a on this exchange. Adjusted share will be automatically set to 0% as described in . You can observe it on Base coin base coin item 3.2.1 of the methodology Bitcoin page Did we miss something you’d like to know? 5. Conclusion Most if not all crypto price aggregators use incorrect and nontransparent price calculation methodology We propose a step by step methodology that has no grey areas and yields reliable average prices All inputs and outputs used in our methodology are readily available on THE SILICOIN With this article we aim to increase transparency of the crypto space. It is the first article in the series devoted to financials of cryptocurrency ecosystem. Parts 2 and 3 will follow. We’d like to reiterate the advice we gave in our post about — be critical about all information you consume. cryptocurrency news sources Open discussion is the best way to achieve consensus. Please address your questions and express your thoughts in the comments below. Transparently yours, team THE SILICOIN