Should you be trading Bitcoin? – The Programming Prodigy

Should you be trading Bitcoin? – The Programming Prodigy

Many of you may remember the cryptocurrency boom in 2017, sending the prices of various virtual coins soaring to unexplainable prices. Looking historically at the price of some cryptocurrencies, there is high volatility[1], hence the asset tends to fluctuate quite often. This is of course quite beneficial for investors, as a short term investment, large sums of profit can be made from trading these assets.

The trading of cryptocurrencies involves the exchange between a cryptocurrency and a token value such as BTC to USDT. USDT is a Tether, which is a blockchain-based cryptocurrency whose crypto coins in circulation are backed by an equivalent amount of traditional fiat currencies. 1 USDT = 1 US Dollar. To gain a profit whilst trading would involve the purchasing of an asset when it is low in price and selling it at a higher price and to do this efficiently and effectively, one has to implement a trading strategy. It is important to note that when choosing an asset to trade, a high trading volume is crucial [2] as it means that there are people willing to trade with you, and therefore you can obtain your desired position at a low cost.

A trading strategy is an algorithmic plan that aims to create a profitable return. There are a range of strategies that can be used, however, this article will investigate whether mean reversion would be suitable and if not, what would be more optimal. Before I explain what mean reversion is, I’d like to outline a few key definitions. Mean is the average price of a given set of data, and the standard deviation is how far the data spreads from the mean. Mean Reversion is the assumption that the price of an asset will fluctuate around the mean, tending to move to an average price over time; suggesting that if a price is a number of standard deviations away, it is expected to return to the mean price at that point. Hence, the asset is purchased when below the mean, and sold when above the mean, at these standard deviations [3]. Conversely, an asset might mean avert, which suggests the price keeps rising or dropping at the point where it is expected to be returning to the mean. This may be due to the overreaction or fear of investors, during periods such as a boom, or surge of investors into an asset [4]. If an asset possesses mean averting properties, a momentum strategy can be implemented as the suggestion that an asset will keep rising for a longer period can be used to generate returns. Doing some background research, there is evidence to suggest that the price of bitcoin does not mean revert, but in fact mean averts [5]. There is evidence that during World War II and the Depression, assets experienced mean reversion, however, it is suggested that evidence for mean reversion is an overstatement, as taking into account these periods are inaccurate [6].

Utilizing Python and its various statistical libraries I aim to experiment and model the trading of BTC and ETH, which are 2 of the most popular and high volume traded cryptocurrencies. Data obtained from Binance, one of the largest cryptocurrency trading platforms, would be used when testing my hypothesis, during the periods of 2017, 2018 and 2019. Initially, the Augmented Dicky-Fuller Test (ADFT) would be used to test for mean reversion. ADFT tests the null hypothesis that a unit root is present, and the alternative hypothesis being there is no unit root, suggesting the time series is stationary. A time series is stationary if a single shift in time doesn’t change the time series statistical properties, in which case a unit root does not exist [7]. If the P-value given by the test is below the significance level, we can reject the null hypothesis, as the is insufficient evidence to suggest a unit root is present. This test can easily identify where a data set experiences mean reversion, as a stationary time series will be mean reverting, hence have no unit-roots. If there is evidence to suggest mean reversion, a strategy based on this principle, using the concept mentioned earlier would be ideal; purchasing the asset when it is a number of standard deviations below the mean, and selling the asset when it is above the mean.

Alternatively, if the ADFT suggest that there is sufficient evidence to accept the null hypothesis, that would mean that the asset is experiencing mean averting properties. If this is the case then a strategy based on the momentum principle would be ideal. This is because there is evidence that the asset’s price will continue rising/falling, even when several standard deviations away. This means that for a given period of time, days or hours if the price of an asset is above the rolling mean it is considered attractive for purchase, and when it is below the mean, it is unattractive hence should be sold; as we would be expecting a further drop at this point.

Whist testing, the data will be split into two groups. Group 1 will be the fit data or historical data, used to create a strategy and fit it to the data. Group 2 will be the out of sample data, which the strategy will be tested on to see if it gives positive or negative results. The out of sample data simulates what would happen when deploying the strategy in real life. Sometimes a strategy may be overfitted to the historical data, meaning it works amazingly with historical data giving large returns, however, it works so perfectly with the historical data, that it no longer works with the out of sample data [8]. The fit data or historical data is used to create, test and optimise the strategy, to gain a profitable return.

During the fit-data testing, two factors were altered: commission fees and the rolling mean window, to generate a number of tables for daily and hourly data for the cryptocurrencies being investigated. Altering the rolling-mean window changes the number of previous prices we use for constructing the moving average. Different cryptocurrency trading platforms have different commission fees, hence by investigating the results at a range of fee levels, this will simulate a more realistic scenario. I aim to experiment with fee levels of (0.01, 0.05, 0.1, 0.5,1%) [9] and rolling windows of (2,5,7,10,15,20,24) days or hours. Identifying the best performing parameters on the historical data (2017–2017), it would then be used and attempted on the out of sample data (2019).

Testing with the ADFT on data from 2017 and 2018, at the 1,5 and 10% significance levels, the P-value was greater than all the values given at the significance levels. This suggests that there is sufficient evidence to accept the null hypothesis which states that there is a unit root. As mentioned before, the presence of a unit root means the data is not a stationary time series, hence suggesting it doesn’t experience mean-reverting behaviours. This is most likely due to the number of spikes in 2017 and 2018 that were caused by the crypto-boom, where lots of people were investing in the currencies to try to make profits. However, what this does identify is that the prices of cryptocurrencies tend to continue going in the same direction once they move, rather than returning to the mean. As a result, it would be more appropriate to use a momentum-based strategy.

A momentum-based strategy is based on the assumption that once an asset’s price has moved in a certain direction, it will keep moving in the same direction for a period of time. If the price of an asset is above the mean then we can expect it to keep rising, and so we would purchase the asset with the aim of selling it at a high price later. The opposite being done if the price of an asset is below the mean. In addition, it is possible to limit the number of tokens at any given time, e.g. to a maximum of 5 to reduce the amount of risk being held. For example, if we had 100 tokens, this poses a high risk as a sudden drop in price can cause our overall loss to be increased to a very significant amount.

At the green circles, we would purchase the asset, and at the red circles, we would sell the asset. Utilizing this principle, this strategy will be run on historical hourly data, during 2017 and 2018, for the trade of BTC for USDT; generating a table of profit and loss, by varying the parameters of rolling windows for the moving average (2, 5, 7, 10, 15, 20, 24 hours) and different fee levels (0.01, 0.05, 0.1, 0.5, 1 %).

Published at Mon, 18 Nov 2019 23:14:06 +0000

{flickr|100|campaign}