Gold Price Forecast: A Summary of the Integration of Economic

Factors and Calculation Methods

Bowen Xie

East Los Angeles College, 1301 Avenida Cesar Chavez, Monterey Park, U.S.A.

Keywords: Gold Price, Machine Learning, Deep Learning, Economic Factors.

Abstract: As an essential safe-haven asset and investment tool in the global financial markets, the price movement of

gold has been widely watched. The price of gold is affected by various economic factors, making its prediction

challenging. In recent years, with the development of machine learning and deep learning technology, scholars

have begun combining traditional econometric models with advanced computational models for gold price

prediction to improve prediction accuracy and provide a reference for investment decisions. This paper

synthesizes the research in gold price forecasting from the perspective of economics and computational

methods. Based on the background of the gold market and the factors affecting the price, it compares the

progress of the application of traditional time series models and machine learning models, discusses the

performance, advantages, and disadvantages of the different models, and finally puts forward the challenges

faced by the current research and the direction of future development. Several studies have shown that

machine learning models incorporating economic factors have yielded promising gold price prediction results.

This can better capture the nonlinear fluctuation characteristics of gold prices and provide valuable references

for the investment market.

1 INTRODUCTION

Gold plays a pivotal role in the global financial

system, as an essential jewelry and industrial material

and as a key component of central banks' reserves and

investors' asset allocation. Because of its stable value

and high resistance to inflation, gold has always been

seen as the backbone of the fight against financial

market uncertainty. Studies have shown that the price

of gold is closely related to the macroeconomic

situation: for example, gold is often used as a hedge

against stock market risks, against inflation, and as a

"safe-haven" in times of currency depreciation or

financial crisis, with its price showing a negative

correlation with significant currencies and stock

markets. However, the volatility of the gold price is

also influenced by speculative trading and market

sentiment, which can be unpredictable and drastic.

This price volatility poses risks for investors and

policymakers and highlights the importance of

accurate gold price forecasting.

Gold price forecasts rely heavily on econometric

https://orcid.org/0009-0006-3560-5386

methods and time series models. For example, the

Autoregressive Integrated Moving Average (ARIMA)

model is often combined with historical price trends

to make forecasts, which, to a certain extent, captures

the cyclical changes and trends in gold prices. At the

same time, researchers also consider the impact of

macroeconomic variables on gold prices by

incorporating factors such as exchange rates, oil

prices, and inflation rates into regression models or

vector autoregression (VAR) models to improve the

persuasiveness of their forecasts. However, such

traditional models are often affected by their

limitations when dealing with sudden situations in

financial markets.

The good news is that in recent years, machine

learning (ML) and deep learning (DL) methods have

excelled in time-series forecasting in finance and

have gradually begun to be applied in the gold price

forecasting field. Rather than relying on purely

statistical models, these data-driven methods are able

to automatically understand complex patterns in large

amounts of historical data while adapting to their

nonlinear relationships. For example, machine

360

Xie, B.

Gold Price Forecast: A Summary of the Integration of Economic Factors and Calculation Methods.

DOI: 10.5220/0013697400004670

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 2nd Inter national Conference on Data Science and Engineering (ICDSE 2025), pages 360-365

ISBN: 978-989-758-765-8

learning models such as Support Vector Regression

(SVR) and Random Forest have demonstrated

superior performance in gold price prediction. At the

same time, deep learning models such as Long Short-

Term Memory (LSTM) can capture the time series'

long- and short-term dependence, which also

significantly improves the forecasting accuracy of

financial series, including the price of gold. In this

context, combining insights from economics with

advanced computational models would be the best

option to improve the accuracy of gold price

forecasting. The remainder of this review will

introduce the main forecasting methods and their

applied research progress, analyze the effects of

different models and suggest future research

directions.

2 SUMMARY OF PREDICTION

METHODS

Gold price forecasting methods can be broadly

categorized into two types: traditional time series and

econometric methods and machine learning and deep

learning methods. The former represents an economic

perspective, mainly modeling with historical patterns

and considering macro factors; the latter reflects a

computational perspective, mainly learning

forecasting patterns from data. The following will

summarize these two types of methods and related

research separately.

2.1 Traditional Measurement and Time

Series Model

The first is time series modeling; as one of the most

widely used traditional methods, the ARIMA model

is often used to capture the autocorrelation structure

in time series data. Many studies have used ARIMA

as a benchmark model to forecast gold prices. For

example, Yang used the ARIMA model to model the

forecasting trend of gold prices with some accuracy

(Yang, 2019). This model works well for linear and

smooth time series, but the actual gold price series

tends to be nonlinear and unstable, which requires

differencing and parameter tuning of the model to

accommodate this complex pattern. In addition, many

studies have combined ARIMA with other methods,

such as using ARIMA to forecast long-term trends

and then mixing it with Support Vector Machines

(SVMs) to capture the nonlinear portion of the

residuals, which can serve to improve forecasting

performance.

The second is the economic factor regression

model; in addition to the pure time series model,

researchers also often use multiple regression models

to incorporate macroeconomic variables into gold

price forecasting; typical examples include Ismail and

other scholars as early as 2009 on the application of

multiple linear regression in gold price forecasting. In

fact, many common economic indicators affect the

price of gold, such as the price of crude oil, inflation

rate, stock market index, and so on. By regressing

these variables with gold price, the fluctuation of gold

price can be explained from the perspective of

economic fundamentals. However, such linear

regression makes it challenging to characterize the

more complex nonlinear relationship between

variables. For this reason, using more advanced

econometric models, such as GARCH-type models,

to describe the volatility dynamics of the gold price

or using structural equations and cointegration

analyses to explore the long-run equilibrium

relationship would be more effective. In general,

however, traditional econometric models do

frequently falter when dealing with highly volatile

and nonlinear financial data.

2.2 Machine Learning Model

In addition to the typical economics methods

described above, machine learning methods are also

well worth using. This approach does not rely on strict

modeling assumptions but instead extracts valuable

features from historical data and applies them to

forecast the price of gold with high accuracy.

Regression trees and integrated models are two of the

most commonly used methods. Decision tree

regression fits nonlinear relationships by recursively

splitting the data and is suitable for determining the

key intervals and thresholds that affect the price of

gold. Random Forest model, such as the decision tree-

based integration method, has a higher level of

robustness and accuracy because of the combination

of the results of multiple trees. Studies have shown

that when dealing with noisy data, Random Forest

tends to achieve more minor mean square error

(RMSE) than the simple linear model and has better

prediction performance. Sharma et al. compared the

effectiveness of various supervised learning models

for predicting the price of diamonds and ultimately

found that Random Forest has the smallest RMSE and

the highest accuracy rate. Also, for the continuous

variable prediction task of gold price, Random Forest

proved to be one of the effective models.

Another machine learning method is SVR, which

is a method that uses kernel function mapping to

Gold Price Forecast: A Summary of the Integration of Economic Factors and Calculation Methods

361

transform the input features into a high-dimensional

space and find the optimal hyperplane in the space to

minimize the prediction error value. SVR is widely

used in financial time series prediction; Yuan et al.

have combined market sentiment analysis and genetic

algorithm optimization with the LS-SVR model to

predict the gold price and achieved high accuracy.

Yuan et al. combined market sentiment analysis and a

genetic algorithm to optimize the LS-SVR model for

gold price forecasting and achieved high accuracy. In

addition to these methods, other algorithms, such as

regularized variants of multiple linear regression

(ridge regression, Lasso regression), k-nearest

neighbor regression (KNN), etc., have also been used

for modeling gold prices. Overall, these machine

learning models can better capture the complex

relationship between gold prices and

multidimensional features than traditional statistical

models and tend to provide better predictions with

sufficient training data.

2.3 Deep Learning Model

Due to the nonlinear pattern of gold price changes,

deep learning methods that automatically learn high-

level features from data through a multilayer neural

network structure have a significant advantage in

predicting in the face of such nonlinear time series.

Specifically, for example, networks like LSTM are a

special kind of recurrent neural network, which is

very well suited to meet the long-range dependence

in processing time series data due to its gating

structure that can store and update long-term states.

In gold price prediction, the LSTM model is able to

utilize the price information of the past ages to predict

future trends, which has a data advantage over the

traditional model that only refers to recent data. A

study comparing the performance of LSTM and

ARIMA models in gold price prediction shows that

LSTM significantly outperforms ARIMA in terms of

RMSE and MAPE, etc. Yurtsever also applies LSTM

and its bidirectional LSTM and GRU variants to gold

price prediction, demonstrating that deep learning

models have higher prediction accuracy than

traditional methods. Prediction accuracy compared to

conventional methods (Yurtsever, 2021).

There is also a beneficial model called a hybrid

neural network model, which is a hybrid model

formed by combining different types of neural

networks to improve prediction performance further.

Among them, Convolutional Neural Network (CNN)

is good at extracting local time-series features from

data, while combining CNN with LSTM (i.e., CNN-

LSTM model) can simultaneously capture the

regional patterns of the gold price series and satisfy

its global time-dependence, which is very effective.

CNN-LSTM model proposed by Livieris et al.

(Livieris et al., 2020). has also achieved good results

in predicting the time series of the gold price. Good

results. Similarly, some studies integrate signal

processing methods such as variational modal

decomposition (EMD) with LSTM, and some even

incorporate attention mechanisms and optimization

algorithms to enhance the model's adaptability to the

dramatic fluctuations of gold prices. In addition, some

studies combine integrated learning with deep

learning, such as using the XGBoost algorithm to

screen features or provide initial predictions and then

fine-tuning the results by neural networks to improve

the robustness of the model. Jabeur et al. utilized the

Extreme Gradient Boosting (XGBoost) algorithm.

They combined it with the SHAP interpretation

method to scientifically predict and interpret the price

of gold, demonstrating the model saturation and the

economic meaning of feature influence (Jabeur et al.,

2021). The financial implications of the model's

characteristic influence are shown.

From the analysis of the above models, they have

their own advantages and disadvantages: traditional

econometric models have good interpretability and

rich economic implications, but it is challenging to

capture the complexity of the nonlinear relationship;

on the contrary, machine learning and deep learning

models can provide higher prediction accuracy, but

often they are regarded as a "black box" due to the

lack of direct economic interpretation. ". Therefore, a

significant trend in recent years has been to integrate

the advantages of both, for example, by incorporating

macroeconomic variables into machine learning

models or by interpreting the decision-making

rationale of complex models, which can be done in a

way that ensures accuracy while improving the

transparency and usefulness of the models.

3 ANALYSIS OF LITERATURE

RESULTS

Since its realization, a large body of literature has

examined and empirically compared the performance

of different models in gold price prediction. The

general trend is that machine learning and deep

learning models have higher prediction accuracy

compared to traditional methods. For example,

Makala and Li compared the prediction of the gold

price by ARIMA and the support vector machine.

They found that the error of price prediction was

ICDSE 2025 - The International Conference on Data Science and Engineering

362

reduced after introducing a nonlinear machine

learning model (Makala and Li, 2021). Tripurana et

al. reviewed the effectiveness of various machine

learning algorithms for gold price prediction,

including linear regression, decision trees, random

forests, SVR, and neural networks, emphasizing the

importance of integrating multiple economic factors

and feature engineering (Tripurana, 2022). Their

study shows that machine learning models are able to

predict gold price movements more accurately when

relevant macro variables are taken into account.

Let's look at the deep learning modeling side of

the equation, where the results of many studies have

demonstrated superior performance over both

traditional and shallow models. A survey by Yurtsever

showed that LSTM models already outperform

classical methods and that combining bidirectional

LSTMs or gated recurrent units (GRUs) can further

improve the accuracy (Yurtsever, 2021). Another

study utilized a combination of LSTM and fuzzy

systems to predict the price of gold, again obtaining

results below the error of the traditional approach.

Livieris et al.'s hybrid CNN-LSTM model excelled in

long-term prediction, with experiments on monthly

gold price data achieving high goodness of fit

(Livieris et al., 2020). Taking the metrics together,

deep learning models typically excel in measures of

prediction accuracy such as mean squared error (MSE

or RMSE), while in directional prediction (e.g., up

and down direction), some machine learning models

such as classification trees also achieve good results

like Sadorsky uses a random forest classifier to

predict the up and down direction of the price of gold

and silver, with an accuracy rate of 1.5 percent

compared with that of randomized predictions

(Sadorsky, 2021)

Nonetheless, some studies point out that the

amount of data and parameter selection influences the

advantage of deep learning models. If the training

data is insufficient or the parameters are not tuned

properly, the complex model may not always win.

Therefore, scholars emphasize the importance of

comparing benchmark models, which should be

comprehensively judged by cross-validation and

multi-indicator assessment. In addition, considering

that gold prices are significantly affected by

unexpected events, such as financial crises or public

health events (e.g., the New Crown Epidemic) that

can lead to inaccuracy of the model's prediction based

on historical patterns, some studies have specifically

improved the model for gold price prediction in times

of epidemics. For example, Mohtasham Khani et al.

explored a methodology that combines epidemic-

related variables and deep learning in the presence of

epidemic shocks to improve the reliability of gold

price forecasts during special periods (Mohtasham

Khani et al., 2021).

Overall, today's literature generally endorses the

application of machine learning, intense learning, to

gold price forecasting but also emphasizes the

importance of incorporating economic theory and

adequate data. Precisely because multi-model

integration and hybrid approaches tend to achieve

more robust results, some recent studies have

attempted to introduce explanatory methods (e.g.,

SHAP values) to reveal the factors that the models

consider most essential to influence the forecasts,

thus making them more informative.

4 CHALLENGES AND

PROSPECTS

Although researchers have made significant progress

in the field of gold price prediction, there are still

many challenges in further research. First, regarding

data and feature selection, the gold price is still

affected by multiple factors, and how to select and

extract key features is still inconclusive. Future

research can explore how to introduce more real-time

data sources, such as news sentiment analysis,

Internet search trends, etc., to enrich the model's input.

Second, the generalization and robustness of the

model remain a problem. Financial markets are often

unstable, and the fact that a model performs well on

historical data does not guarantee that it will remain

valid in the future. Therefore, there is a need to

develop adaptive forecasting models that can be

updated promptly based on real-time data. Not only

that, but the impact of abnormal events (black swan

events) on the gold price is also often beyond the

scope of experience accumulated by the model.

Improving the model's ability to respond to extreme

situations is a direction worthy of research. This can

be done by combining scenario analysis or building a

hybrid forecasting system that relies on data-driven

models during smooth periods while combining rules

or expert judgment during abnormal periods.

At a time when the interpretability of models is

increasingly emphasized, it is more important for

investors to understand the reasons why models give

predictions rather than being satisfied with highly

accurate predictions. Future research could make

greater use of explainable artificial intelligence (XAI)

methods to explain the decision-making process of

complex models. Use feature importance analysis to

clarify which macro indicators have the highest

Gold Price Forecast: A Summary of the Integration of Economic Factors and Calculation Methods

363

weight in model predictions or use locally

interpretable models to analyze the drivers of

predictions at specific points in time. This will

facilitate the translation of forecasts into the market

insight process.

Interdisciplinary integration will be an essential

development trend in gold price forecasting. Gold

price movement is not only a financial phenomenon

but is also affected by multiple factors, such as

politics and society. If the theories of financial

economics are combined with machine learning

techniques, new forecasting frameworks can be

generated. Just like the idea of integrating game

theory or multi-agent simulation, a complementary

effect can be achieved with data-driven models. In

addition, with the increasing computational power,

the realization of real-time prediction and automated

trading is no longer impossible, and future research

can further embed the prediction models into trading

decision-making systems so as to assess their actual

returns and risks.

5 CONCLUSIONS

In general, gold price forecasting is a valuable and

challenging research topic in the field of finance. This

paper summarizes the significant research advances

in this area from both economic and computational

methodological perspectives. Both traditional

econometric models, which provide the underlying

framework and economic explanations for gold price

forecasting, and machine learning and deep learning

models, which are capable of handling large-scale

historical data with high accuracy, play a crucial role.

The approach of incorporating macroeconomic

factors into the forecasting models, in turn, takes the

performance of the models to the next level and also

enhances their real-world applicability, providing

more reliable theoretical and practical support for

gold price forecasting.

However, even with modeling, there are still many

sources of uncertainty in accurately predicting the

price of gold, not to mention that the applicability of

models needs to be tested in a changing market. In

order to improve the reliability of forecasts, future

research directions need to focus on the

generalization ability of models and the adaptation to

abnormal situations, which may require the

integration of multi-model approaches and the

accumulation of experts' experience. At the same time,

enhancing the interpretability of model results is

crucial for the application of forecasts in practical

decision-making. All in all, gold price forecasting

should be a cross-disciplinary collaborative process

that combines economic insights with data science

techniques, utilizing machine learning models to

improve forecasting accuracy on one hand and

financial theories to explain the results and guide

model improvement on the other. Only in this way

can forecasting research provide investors and

policymakers with truly valuable guidance on how to

respond to the rapidly changing gold market.

REFERENCES

Arouri, M. E. H., Lahiani, A., & Nguyen, D. K. 2015.

World gold prices and stock returns in China:

Insights for hedging and diversification strategies.

Economic Modelling, 44, 273-282.

Bampinas, G., & Panagiotidis, T. 2015. Are gold and

silver a hedge against inflation? A two-century

perspective. International Review of Financial

Analysis, 41, 267-276.

Jabeur, S. B., Mefteh-Wali, S., & Viviani, J.-L. 2021.

Forecasting gold price with the XGBoost

algorithm and SHAP interaction values. Annals of

Operations Research.

https://doi.org/10.1007/s10479-021-04233-1

Livieris, I. E., Pintelas, E., & Pintelas, P. 2020. A

CNN–LSTM model for gold price time-series

forecasting. Neural Computing and Applications,

32(23), 17351-17360.

Madziwa, L., Pillalamarry, M., & Chatterjee, S. 2022.

Gold price forecasting using a multivariate

stochastic model. Resources Policy, 76, 102544.

Makala, D., & Li, Z. 2021. Prediction of gold price

with ARIMA and SVM. Journal of Physics:

Conference Series, 1767(1), 012022.

https://doi.org/10.1088/1742-

6596/1767/1/012022

Nguyen, Q. N., Bedoui, R., Majdoub, N., Guesmi, K.,

& Chevallier, J. 2020. Hedging and safe-haven

characteristics of gold against currencies: A

multivariate dynamic copula approach. Resources

Policy, 68, 101766.

Sadorsky, P. 2021. Predicting gold and silver price

direction using tree-based classifiers. Journal of

Risk and Financial Management, 14(5), 198.

Tripurana, N., Kar, B., Chakravarty, S., Paikaray, B.

K., & Satpathy, S. 2022. Gold price prediction

using machine learning techniques. In

Proceedings of ACI@ISIC 2022: Advances in

Computational Intelligence Workshop (pp. 274-

281). CEUR-WS.

Yuan, F.-C., Lee, C.-H., & Chiu, C. 2020. Using

market sentiment analysis and genetic algorithm-

ICDSE 2025 - The International Conference on Data Science and Engineering

364

based least squares support vector regression to

predict gold prices. International Journal of

Computational Intelligence Systems, 13(1), 234-

246.

Yurtsever, M. 2021. Gold price forecasting using

LSTM, Bi-LSTM, and GRU. European Journal of

Science and Technology, 21, 1-8.

Gold Price Forecast: A Summary of the Integration of Economic Factors and Calculation Methods

365