Data Analysis of M2 Money Supply

The analysis encompasses several general areas like, loading and preprocessing the data, visualizing
the temporal trends and monthly distributions, calculating descriptive statistics and daily variations,
and the use of outlier detection methods. These steps are important in ascertaining how the M2 money
supply behaves and finding potential trends or irregularities in the data being analysed.

Data Loading and Preprocessing:

The very first step of the analysis process was to create the necessary computational
environment. This started with setting up relevant Python packages that are crucially
important in data science and time series analysis. In particular, Pandas was used for the
powerful data structures of Data Frames and data manipulating tools. NumPy was included
for faster computation of numbers, Matplotlib and Seaborn for generating language
informative graphs and graphics, and finally, SciPy and Statsmodels for performing a
statistical computation. With the libraries already loaded, the M2SL data was read into a
Pandas Data Frame form from a CSV format file labelled as M2SL.csv. The Data Frame
formed the main data structure for all the operations that were to follow. Rather interestingly,
the data frame had two columns that were the central focus: ‘DATE’ and ‘M2SL’. As the
‘DATE’ column was an object consisting string representations, further great attention was
put into pre-processing. The pd.to_datetime() function was used to change the format of the
‘DATE’ column from an object to a datetime format. This conversion is essential in time
series analysis because it allows Pandas to see and treat the data as time series data and
therefore a lot of time related features may be accessed. In addition, to improve efficiency for
operations concerning the date, the DATE column was used as the Data Frame index by
issuing the command line df.set_index(‘DATE’,inplace=True). This transformation
significantly streamlines subsequent time series operations.

Time Series Visualization:

A line plot was made to obtain a first perception on how the M2SL evolves in time. This
visualization highlights the general trend of the money supply over the length of the dataset.
The plot was structured with 'DATE' on the x-axis and 'M2SL' on the y-axis, allowing for a
clear visual representation the ups and downs of the money supply over time. From this, you
could create a plot that clearly showed a high-level increase in the money supply over time.
In particular, a larger and faster increase was observed in the later parts of the dataset,
indicative of a change in monetary policy or economic condition.

Monthly Distribution Analysis:

In order to explain the differences of M2SL between the months, the authors utilized Seaborn
to construct a box plot. It is robust when it comes to displaying the values of a particular
variable spread out across categories which in this case are the months of the year. For the
purpose of this analysis, the DATE column was used to generate a new column called Month
using the accessor .dt.month_name(). This column contained the names of months in full and
as a result enabled efficient data grouping. The box plot illustrated the M2SL measures of
central tendency and dispersion together with the extreme values for the distribution by
months. From the findings of the box plot, considering the circulation of M2SL on a monthly
time frame, there is not a big difference in M2SL in most months. The characteristic of the
box indicated the absence of some M2SL outliers in certain months, more specifically the
ones with a higher or lower than expected M2SL values for that month.

Descriptive Statistics and Daily Variation:

In order to add to the more quantitative comprehension of the M2SL data, a range of
descriptive statistics was computed. This included the calculation of mean, median and
standard deviation. It was determined that the mean (5389.55) value of M2SL over the period
addressed in the data as average, while the figure for middle or dominant value, median
(3368.85) through its characteristics is less prone to outliers. Moreover, the significant
disparity that exists between the mean and median is indicative of a right skewness in
distribution which simply means that there were higher values in the data that tended to pull
the mean up. It was established in this case that the standard deviation (5717.90)
quantitatively assesses the degree of disparity or scatter of the data values around the average
point. Also, it confirmed that there was a wide range of values within the data. Additionally,
the .diff() method was also used to find the daily fluctuation in the M2SL. This method works
by determining the value of a data point and the value of the ‘preceding’ data and thus the
difference of the two gives the fluctuation in value for that day. The composite of the average
of the daily fluctuations standing around 26.65 units was an accurate representation of the
scale of the volatility of the M2SL on a daily basis.

Outlier Detection:
In order to detect possible anomalies in the M2SL data, the approach of Z-score was applied.
For each data point, the Z score was determined using the function stats.zscore provided by
SciPy. The Z score is a measure of how far away from the mean a data point is expressed in
terms of standard deviation. To disregard the direction of the deviation, only the absolute
value of the Z-scores was of interest, which allowed focusing on the deviation’s size. A
predetermined cutoff of 3 was used, such that any point with an absolute Z score greater than
or equal to 3 was considered to be an extreme outlier. However, in the dataset which was
worked on, this condition was not satisfied for any point which means that the mean was not
more than 3 standard deviations from the highest value. This does not connote that there are
no extreme values at all but shows that the extreme values set did not go near the set limit.

Further Analysis:
All future analysis can progress on this robust groundwork. The following suggestions could
be enacted in turn:

1. Time Series Decomposition: It would be useful to know how the M2SL is structured by
looking at the components (trend, seasonality and residual) that drive it and this could be
achieved through decomposition of the series.
2. Statistical Tests: Applying such tests as the Augmented Dickey-Fuller Test would assess
the stability of the time specific data which seems to be quite important for the majority
of time series forecasting methods.
3. Forecasting: Time series techniques can be developed and tested so as to come up with
estimates or predictions of values of M2SL over time for economic or policy planning
purposes.
4. Sensitivity Analysis of Outlier Detection: The use of varying Z-scores for detection of
outliers would explain to a greater extent outliers and their importance in the data
analysed.
5. Investigating External Factors: An investigation of the other primary or macroeconomic
variables related to the M2SL can help understand the determinants of money supply.
Prophet
This report outlines the results of a time series forecasting model performed on the historical data of
the monetary aggregate (M2SL). The period under consideration stretches from January 1961 through
October 2020 providing adequate period for analysis. The Prophet model complements the analysis
by improving the usability of the data. With this model, time period is extended by 24 months from
October 2024 to October 2026 in which the monetary aggregate values are estimated. The aim of this
forecasting exercise is to give insights into the long term prospects, seasonal activities and growth of
the monetary aggregate towards aiding economic development and planning.

The results of the Prophet model across various locations and time demonstrate a number of important
features of the monetary aggregate’s economics and econometrics data and trend including a
deterministic upward trend and regular cycle. This analysis utilizes the best features of the Prophet
framework which is particularly effective for modelling the time series with strong trend and long
yearly cycles. This report it is reasonable to predict and explain the amount of money in circulation
using this model since it incorporates both estimates and forecasts.

Historical Trends and Model Fit

Overall, the Prophet model has been effective in capturing the upward trend in the monetary
aggregate which has been prevalent during the entire period in question. This trend is a hallmark of a
growing economy where, in the long run, monetary aggregates increase due to inflation, increased
liquidity, and macroeconomic growth. In this case, the Prophet model performed excellently, as it was
able to fit with historical data from 1961 to 2024.

One notable strength allowing the model to perform well is its ability to control for yearly seasonality
which is apparent in the data. It is normal for an economy to exhibit seasonal patterns in monetary
aggregates owing to economic cycles, fiscal policies and business activities. The fit of the model
suggests that yearly seasonality is satisfactorily captured in the data resulting in episodes of moderate
growth around specific months and business activities.

The model thus created, while targeting long run trends and yearly seasonality, is deigned purposely
to leave out daily and weekly seasonality. This is the case since monthly data does not require such
fine detail. In conclusion, the historical fit corroborates and extends the Prophet framework reliability
in examining monetary aggregate trends across large periods.

Forecast Results (November 2024–October 2026)

The forecast rendered by Prophet model illustrates the upward trend uncovered from historical data.
Over the span of 24 months, the forecast suggests that the monetary aggregate would grow steadily
without any major changes from the previously observed patterns. This likely is premised on the
expectation that during the forecast horizon there will be no shocks to the economy and there will also
be no changes in structural trends that have been observed historically.

The monetary aggregate is anticipated to increase by October 2026 to about 23,930.37, with the 95%
confidence interval of 23,242.86 and 24,592.50. As the forecast extends further in time, so does the
confidence interval, this is as a result of predicting long term economic trends which often are faced
with great degree of uncertainty. Nonetheless, the intervals are relatively tight suggesting that there is
a great deal of confidence in the model results.

The results of the forecast demonstrate the degree of robustness and coordination of the monetary
aggregate across time and space. The stability characterizing the sequence is consistent with the
presence of some fundamental economic strength and the fact that there were no significant
destabilizing shocks in the historical data. All the changes in the anticipated growth path which the
model hypothesized are consistent with the observed performance and offer support for planning and
policy making purposes.

Key Observations
By analyzing the forecast results, the behavior of the monetary aggregate and its development may
have some form of future direction. The first piece of information to be extracted from the data is the
fact that the data assumes a monotonic trend, no doubt arising due to macroeconomic considerations
like rising inflation, increased liquidity, and shifts within the financial system. That trend makes it
clear how much long term economic planning is needed because the expansion of monetary
aggregates may have dire consequences for inflation, interest rates, and the economy’s welfare
balance.
The second aspect is the prediction of the monetary aggregate. The fuzzy sets that accompany
prediction contain rather small intervals and that goes to show how stable the time series is and how
well the Prophet model is able to represent its dynamics in the first place. Such stability gives a firm
base for economic agents expecting changes in economic conditions in the foreseeable future.

Finally, the uncertainty contained in any forecast, as which every model possesses, deserves
recognition. The model pays due heed to the historical aspect but does not tend to factors that lie
entirely outside it, for instance, economic shocks, modified policies, or international relationships.
Such variables might enhance the level of forecast error and therefore be factors to consider in the
interpretation of the hypothetical results.

Limitations
Although the Prophet model is efficient in predicting future events, it has some limitations, and its
usage is subject to specific scenarios. One of them is relying on past historical trends and patterns to
hold true in the future. In other words, the model assumes that a growth path that has been witnessed
in the past will still hold true in periods to come, which is highly unlikely in times of huge economic
volatility or drastic policy shifts.

Another limitation is the lack of attention on short range seasonality. It is true that Prophet model
captures annually recurring trends, however, daily or even weekly trends which might be important
for highly disaggregated datasets do not get included. However, for datasets in month frequency, this
limitation is not as major.

The above mentioned limitations pertain to the broad nature of the Prophet forecast and consequently
reinforce the need for informing of the model’s findings with other forms of analysis such as scenario
analysis and alternative model approaches.
Discussion

This study on the U.S. M2 money supply provided comprehensive insights into the patterns
and dynamics of monetary growth through the use of traditional statistical models and
advanced machine learning techniques. The findings offer significant implications for various
stakeholders, including policymakers, economists, and analysts, by presenting a detailed
understanding of historical trends and reliable forecasts for future economic planning. The
integration of different methodologies underscored the strengths and limitations of each
approach, offering a broader perspective on time series forecasting.

The initial phase of the analysis emphasized the importance of data preparation. Stationarity,
a critical assumption for many time series models, was evaluated using the Augmented
Dickey-Fuller (ADF) test. The test confirmed that the original data lacked stationarity,
necessitating transformations such as differencing and logarithmic adjustments. These
preprocessing techniques successfully addressed the issue, ensuring that the data was
appropriate for forecasting. This step also highlighted the fundamental role of data
transformation in producing accurate and meaningful predictions, especially in economic
datasets prone to trends and seasonality.

The forecasts across all models revealed a consistent upward trajectory in the M2 money
supply, indicating ongoing monetary expansion in the U.S. economy. This finding is
particularly relevant given the potential implications for inflation, liquidity, and broader
economic stability. A continuous rise in money supply often serves as a double-edged sword,
providing liquidity to fuel economic growth while simultaneously posing risks of inflationary
pressures if left unchecked. For policymakers, this insight reinforces the need for proactive
measures to strike a balance between monetary expansion and price stability. By leveraging
these forecasts, central banks and financial regulators can make informed decisions regarding
interest rates, open market operations, and other tools of monetary policy.

The comparison between traditional statistical models and machine learning approaches
revealed distinct strengths and weaknesses. The ARIMA and SARIMA models, which have
long been established as reliable tools for time series forecasting, performed well in capturing
linear relationships and seasonal patterns. Their simplicity and interpretability make them
valuable for understanding straightforward trends in economic data. However, their
limitations became evident when faced with more complex, non-linear dynamics that
characterize real-world monetary data. These traditional models are constrained by their
reliance on pre-specified assumptions and parameters, which may not fully capture the
underlying intricacies of the data.

On the other hand, machine learning models, particularly XGBoost and Long Short-Term
Memory (LSTM), demonstrated exceptional accuracy and adaptability. XGBoost excelled in
capturing non-linear patterns and interactions within the dataset, owing to its ability to handle
complex feature interactions and missing data. Its high performance in this study showcases
its suitability for economic forecasting, where data often exhibits irregular and intricate
behaviors. The LSTM model, designed specifically for sequential data, emerged as the most
precise tool for forecasting. By leveraging its capability to learn long-term dependencies,
LSTM effectively captured both short-term fluctuations and broader trends in the M2 money
supply. This adaptability underscores the transformative potential of deep learning techniques
in time series analysis, especially when dealing with data that exhibit irregularities or multi-
dimensional patterns.

The implications of this study extend beyond theoretical findings, offering practical value for
a wide range of stakeholders. For policymakers, the projected increase in the money supply
provides a basis for designing monetary policies aimed at mitigating potential inflationary
risks while supporting economic growth. Economists and financial analysts can utilize these
insights to refine their understanding of liquidity dynamics and macroeconomic trajectories,
thereby enhancing their ability to predict and respond to shifts in economic conditions.
Moreover, the juxtaposition of traditional models with advanced machine learning
approaches highlights the importance of adopting a hybrid analytical framework. By
combining the interpretability of statistical models with the predictive power of machine
learning, stakeholders can achieve a more holistic understanding of economic trends.

Despite the promising results, this analysis also brought to light certain limitations that
warrant further exploration. The study primarily relied on historical M2 money supply data,
without integrating external variables such as fiscal policies, geopolitical events, or global
economic shocks. Incorporating these factors in future research could provide a more
nuanced understanding of the drivers influencing monetary trends. Additionally, while
machine learning models demonstrated superior accuracy, their interpretability remains a
challenge. Efforts to integrate explainable AI (XAI) methodologies could bridge this gap,
enabling stakeholders to better understand the mechanisms underlying machine learning
predictions. Furthermore, the findings are specific to the U.S. economy, and replicating this
analysis across other regions or countries could offer comparative insights into global
monetary trends and policy implications.

Overall, the study underscores the importance of leveraging diverse methodologies to address
the multifaceted nature of time series forecasting. Traditional models provide a robust
foundation for understanding linear patterns and seasonal behaviors, while machine learning
models unlock the potential to capture complex, non-linear dynamics. The integration of
these approaches not only enhances forecasting accuracy but also equips stakeholders with a
comprehensive toolkit for informed decision-making in the realm of economic and financial
analysis.