Why we should be wary of ‘expert’ price forecasts

Being such a young market, ~95% of existing crypto assets were created in the past two years, in the absence of meaningful historical data, narratives, news and events have a stronger impact on prices than any particular price support/resistance levels – though there are narratives given for these technical levels as well – such as the breakeven price for miners, Fibonacci levels, "psychologically significant levels" etc.

"Expert" price predictions not just misleading but dangerous

Picking the price at which bitcoin and the wider market will bottom may be just as illusive as forecasting the price by year-end, and can be just as dangerous.

So wayward have the ‘expert’ price forecasts been for bitcoin over the past year that it should be obvious they have no idea what’s really determining price.

"Many individual investors think that institutional investors dominate the market and that these ‘smart money’ investors have sophisticated models to understand prices—superior knowledge. Little do they know that most institutional investors are, by and large, equally clueless about the level of the market."
__ Prof Robert Shiller, Irrational Exuberance__

The most notable to retire his public opinion is divisive bitcoin permabull Tom Lee of Fundstrat who appeared regularly on CNBC with predictions that bitcoin would hit $25k by the end of 2018. In December he wrote to clients, "Because of the inherent volatility in crypto, we will cease to provide any timeframes for the realization of fair value."

Just some of the examples of Tom Lee’s 2018 price predictions based on limited historical events and price data.

CEO of Bitmex Arthur Hayes at one stage last year predicted $50k by the end of 2018 on the premise "I was bullish and I picked a nice round number with psychological significance, though pared it back several times during the year." Former Goldman Sachs partner turned crypto fund manager Mike Novogratz, forecast $40k by year-end. Both have pared back expectations to ~$3k for an indefinite time horizon.

Owners of exchanges or anyone whose salary is dependent on trading commissions offering price predictions should raise a giant red flag for obvious reasons.

Prediction models that don’t take environmental changes into account

However, Lee still insists that the market price is wrong according to his fair value model which values bitcoin around $13-14k. This sort of rigid faith in a theoretical statistical model to predict the future in the complex domain of the real world has been described as ‘Ludic fallacy‘ – or "basing studies of chance on the narrow world of games and dice".

Although finance is full of laws, formulas and terms borrowed from physics, unlike physics there are no laws or models that hold constant. In the attempt to turn economics into a science (see Econophysics) and create models that accurately predict the economic activities of people, forecasting models rarely factor in structural changes in the human sentiment or emotions that drive decision making.

Bitcoin as micro-economy has many moving parts and forecasting its price is similar to forecasting for the economy at large – it is a relatively simple system of different actors (miners, exchanges, retail investors, users etc) with complex behavior and interactions between them. As in the wider economy it’s impossible to be in possession of all information. Models based on empirical data (in particular limited price data) are also prone to suffering from ‘unobservables’ or extraordinary (Black Swan) events that haven’t happened previously.

Before 2016 global exchange liquidity for bitcoin was miniscule as displayed by the BLX volume data, which shows depth of global exchange order books.

For example, most of bitcoin’s historical price data was generated during a period when there were few major exchanges and one dominated the market, Mt Gox, which handled 70% of global BTC transactions before its implosion in 2014. There were also far fewer fiat onramp services; far fewer participants in the market and very little mainstream awareness. Bitcoin mining was also easier, cheaper and more diverse with fewer alternative currencies to mine as well as competing protocols to proof-of-work.

Trinity of statistical errors

Unlike building models for the natural sciences (such as the weather), financial forecasting models are even more subject to modeling errors as the study of markets is afterall a social science – the study of human behavior.

Financial models also suffer more acutely from what has been described as a trinity of errors in model specification (assuming normal distribution models): choosing an inappropriate functional form (polynomial functions, inverse functions, log functions etc); errors in model parameter estimates; and errors in the model failing to adapt to an ever-changing environment (for asset returns, time is volatility as unforeseen events can suddenly happen).

Despite the "fat-tail" distribution of stock returns being well documented (Black Swan events) many financial models adopt normal distribution, which would imply Black Swan events are rare occurrences. However, they are infrequent, not rare, and the tendency is for new information to come in big lumps infrequently. Consider how many "unforeseen events" have happened in the US between 2000-2010 alone – including 9/11 and the sub-prime mortgage crisis.

In the case of Tom Lee, he does not accept that his fair value model based on the number of active wallets is wrong but that the market’s pricing is irrational. This assumes his model’s parameters about the future are all correct, the correct functional utility of bitcoin to the thousands of alternative currencies one could choose from and assumes no errors in the data.

Institutional investors won’t catch a falling knife

Historically bitcoin and crypto markets have relied on the novice retail trader (or "dumb money") to jump in and kick-start a bull run. Now, though, the crypto community is expecting institutional or "smart money" to lift the market, save for the miracle of the SEC granting an ETF in the meantime.

There is little reason apart from anecdote to believe that big banks and institutional traders are waiting to jump in and go long the market and if institutional interest so far is anything to go by (again by inference) it is short interest. According to Federal Reserve researchers, when bitcoin futures contracts were introduced in December 2017, it precipitated bitcoin’s fall from all-time highs as it provided the first opportunity for professional traders to short the asset.

"Smart money" invests early in assets that are cheap and trending upwards and therefore is never likely to step in and catch a falling market such as bitcoin let alone the wider cryptomarket regardless of what they may think is "fair value" – for which there is no right model. It is also a possibility that early adopters and users who have had significant holdings for years through several booms and busts are leaving the market, removing another price floor.

Relationships are complex: Days destroyed, difficulty and price

Like all relationships, things change over time and behaviors that once were predictable breakdown and reverse.

Bitcoin days destroyed (BDD), a gauge for the number of long-held ‘coins’ leaving a wallet and interpreted as BTC moving from holders to speculators, has coincided with significant price action which has proven consistent since 2011.

Aggregate Bitcoin Days Destroyed (BDD), blue line, is interpreted as the movement of "old money" to new speculators. Source: OXT.me

When BDD hit previous all-time highs in July 2011, April 2013, November 2013 and Dec 2017 it preceded a massive rally in price and subsequent bust. In this short historical context, it is unusual that this latest all-time-high in December 2018 came without any subsequent dramatic price movement as BTC has tracked sideways since then (BTC price did fall ~$3,000 between mid- and late November).

This is just one example of extrapolating historical data to predict a future which didn’t eventuate as expected whatever variable has changed to make “this time different” is impossible pinpoint.

The relationship between bitcoin’s price and miners’ operations is the enigma code everyone is trying to crack for an insight to valuing the asset and predicting where it might go.

But bitcoin’s price is trapped in a complex mining dilemma as miners are the natural sellers (along with exchanges) in the market who must sell to cover operation costs while there are no natural buyers apart from exchanges which need to cover withdrawals from time to time. Mining difficulty is a large component of opex for a miner and there is a complex relationship between the two which changes over time.

In a very simplified version, price leads hashrate and hashrate leads difficulty so sharp price rises (for example Dec 2017) can have a profit dampening effect for miners as they lead to more mining competition and an upwards difficulty adjustment (higher expenses). Miners would be prepared for this and be expected to sell BTC stock before the difficulty adjustment (roughly every two weeks or every 2016 blocks) to cover expenses.

Rolling 4-week correlation window of BTC change in price to change in difficulty level.

Looking at the correlation between the change in price and difficulty from June 2017 to June 2018, we can see several extreme positive spikes and immediate large sharp negative reversals (ie. on occasions when price and difficulty both went up it caused an immediate countereffect as price was pushed down by higher difficulty) in the lead up to the bull run of Jan 2018. These extreme positive/negative correlation flips happened on three separate occasions when price briefly doubled to an all-time high within a two-month period and quickly reversed to give up ~80% of the gains.

The first was between late July and September when price more than doubled from ~$2,000 to ~$5,000 and fell back to $2,900. The second spike was Sept-Nov, when price went from ~$3,000 to ~$7,900 only to reverse to ~$5,400; and the last and most sustained was in December when price went parabolic from ~$5,600 to ~$20,000 only to fall $10,000 and try to retest the highs in Jan 2018. It may well have been that it was miners’ selling pressure that contributed to the fleeting all-time-highs and sharp reversals.

The supply-demand dilemma

The precarity of relying on miners to secure the bitcoin network with hashrate was highlighted recently when the hashrate broke violently down through its almost decade-long upward trendline since 2009 and warned about in a recent Bank of International Settlements report.

Hashrate leads the difficulty level and the latter shortly followed also breaking its long-term trendline, dropping on four consecutive adjustment periods as miners either pulled out, went broke or retired machines – indicated by a fall in hashrate.

Looking at the six-month period during which this unprecedented drop occurred (Aug ’18-Jan ’19), we can see the 7-day average number of bitcoins (BTC units) mined, the BLX liquid price of bitcoin in USD and the difficulty/hashrate all moved in tandem. It is unusual behavior for a commodity’s price to move positively with supply and suggests the confidence of miners and the confidence in miners’ ability to secure the network affects the price disproportionately rather than natural demand from buyers in the market.

BTC difficulty level, August – January

The difficulty level and the 7-day average number of BTC mined below shows corresponding movements in August and slight recoveries since then, suggesting the supply of BTC coming into the market does fluctuate corresponding to mining demand.

7-day average of BTC mined in units, August – January

Without any natural demand in the wider market it appears bitcoin’s price is also beholden to mining demand (shown by hashrate) and the variables that determine miners’ breakeven price.

BLX Price, August – January

There is only conjecture as to why this happened, such as unprofitable miners going offline, but the Bank of International Settlements recently published a paper warning of the inherent risks in proof-of-work mining and an impending liquidity crisis in bitcoin when miner’s rewards are diminished without consumer demand replacing mining incentives with transaction fees.

Conclusion

As an investor one should be wary of relying on others’ price predictions for tops and bottoms based purely on historical price and models as there is an entire micro-economy at work in bitcoin and other crypto assets. Without the availability of cash flows and balance sheets that determine fair value for traditional stocks, for example, predicting the value of crypto assets is even more difficult. And without a thorough understanding of the complex economy of relationships at play, crypto price predictions become even more fickle.

Alternative tools and improvements are being explored that look at crypto-networks like social networks instead of traditional assets, using fundamental metrics such as the price to Metcalfe ratio (PMR) and network value to transactions (NVT) ratio.

If cryptocurrencies do mimic the same network effects as social networks, then (as documented by Facebook’s recent troubles and consequent share price tumble) widespread sentiment and narratives changes have the largest effect on price. Given that, using complementary indicators for a turning point in mainstream sentiment (outside the crypto economy) and a bull run should be a better indicator than price.